GENERAL DISCUSSION Dr. B. 3. Mason (Imperial College, London) said: As indicated in my paper, I do not believe that the expansion chamber experiments purporting to observe homogeneous condensation have provided such good confirmation of the Volmer- Becker-Doring theory as is commonly supposed. It seems likely that, because of the great difficulty of preventing nucleus formation by irradiation of metal and rubber surfaces and of the air itself, many writers may have, in fact, observed heterogeneous nucleation. The fact that fair agreement between the theory and experiment is observed at temperatures near 0°C but that serious discrepancies occur at apprecialy lower temperatures suggests that perhaps the former agreement is fortuitous. Ds. W. J. Dunning (Bristol University) said: In reply to Dr.Mason, in my paper I mentioned that a thorough test of the theory would require the experi- mental evaluation of the dependence of droplet concentration and droplet size on the time, the supersaturation ratio and the temperature. A " one shot " ex- periment, as carried out by the cloud chamber method, falls far short of this ideal. Further, there is no unique value for S,,i, which will depend on the experimental arrangements and the skill of the observer. It is perhaps fairer to check the form of the relationship between and other variables as determined by one ob- server, rather than seek exact agreement between observers. I also point out that nucleation theory has been applied in explaining the phenomena which occur in expansion nozzles and wind tunnels; there is the possibility of inverting this approach and using these phenomena as a method of studying nucleation and growth theory.It may be hoped that such an investigation would correspond to an experimentum crucis ; at least more variables would be accessible to measure- ment and control than in the cloud chamber. The other method, in which a jet of vapour is supercooled by mixing with a coolant atmosphere, has distinct promise but has the disadvantage that the theory of turbulent jets does not describe quanti- tatively the experimental data. Dr. E. R. Buckle (Imperial College, London) (communicated): It has been remarked by Dr. Mason that serious discrepancies are found between nucleation rates predicted by theory and those indicated by experiment.The result has been a general lack of confidence in the kinetic theory of nucleation, which never- theless has not been substantially improved for 25 years. I would like to draw attention to three factors which may have contributed to the apparent disagree- ment between theory arid experiment. My remarks are mainly concerned with homogeneous nucleation in condensed systems, particularly the freezing of super- cooled liquids. The theoretical expression of Turnbull and Fisher 1 for the specific rate of homo- geneous nucleation in pure liquids may be written J = K exp (-AW/kT), where A@* is the free energy of formation of a critical embryo, or nucleus, and K is a kinetic coefficient given by K = (N,kT/h) exp (- &/kT). Here, NO is the number of molecules in unit volume of liquid, and E is the activation energy for the assimilation of single molecules by a growing crystal embryo.In applying this theory to the calculation of nucleation rates, a relatively small error is automatically introduced due to a mistake in the derivation of (2). It may be 1 Turnbull and Fisher, J. Chem. Physics, 1949, 17, 71. 596Q GENERAL DISCUSSION shown 192 that the expression for K is too large by a factor g*# where g* is the number of molecules in a critical embryo. Values of g* for a wide range of compounds show 3 that nucleation rates predicted by (1) and (2) are consequently too high by one or two orders of magnitude. Since the degree of uncertainty involved in other quantities assumed in calculating K may for some liquids intro- duce errors of similar size, the above correction need mot necessarily result in marked improvement in calculated nucleation rates.A more serious source of error may arise in the method of using theory to interpret critical supercoolings. It is commonly assumed 4-8 as an experimental criterion that the threshold of freezing in an isolated droplet depends on the forma- tion of a single nucleus during the time of observation. This criterion may be expressed as Here, Vis the volume of a droplet, to is the observation time, and Ts is the freezing threshold. The value for I given by (3) is used to calculate either K or A@*, depending on existing knowledge concerning the liquid-solid interfacial free energy 0. In using eqn. (1) for this purpose, the further assumption is made that the threshold nucleus forms under steady-state conditions, i.e., that the relaxation time is exceeded by the observation time.It is shown elsewhere 3 that where (3) is assumed the nucleation time-lag E, defined by I = JV = lito; T= T,. (3) Q(t) = (t-L)J; t + ~ , (4) where Q(t) is the flow of embryos in time t, may not fulfil this requirement. For example, an upper limit to the time lag, applicable where the nucleating system relaxes from an initial distribution comprising only single molecules, is given by L, N No/J. ( 5 ) Since for the observation of steady-state nucleation L1 <to, the criterion (3) implies that iVoVG1, which is completely unrealistic. The difficulty is removed if the freezing criterion (3) is abandoned in favour of one which avoids arbitrary specifica- tion of the threshold nucleation frequency I.An obvious alternative is L- t o , (6) IN NOVit,, (7) which, using (5), leads to a stationary nucleation frequency exceeding that postulated in (3) by as many as ten orders of magnitude. Although (5) probably over-estimates actual time lags in most types of experiment, since it is based on the most unfavourablle initial condition of the system, it is evident that grave errors may still be introduced into estimates of Kusing (3) as the threshold criterion. Finally, in studies where nucleation rates have been obtained by counting the number of droplets crystallizing in a given time under isothermal conditions, it has not always been clearly established that freezing was initiated in each droplet by a single nucleus, although this is assumed in the interpretation of results.It seems that in this way also nucleation frequencies may have been grossly under- estimated. 1 Turnbull, private communication, 1959. 2 Buckle, to be published. 3 Buckle, Nature, 1960, 186, 875. 4 Turnbull, J. Chem. Physics, 1950, 18, 198, 768, 769. 5 Turnbull, J. Appl. Physics, 1950, 21, 1022. 6 Turnbull, J. Chem. Physics, 1952, 20,41 I.. 7 Thomas and Staveley, J. Chem. SOC., 1952, 4569. 8 De Nordwall and Staveley, J. Chem. SOC., 1954,224.GENERAL DISCUSSION 61 To summarize briefly : in tests of the theory of nucleation in supercooled liquids based upon the magnitude of the kinetic coefficient K there is a tendency for cal- culated values to be 10-100 times too large, while " observed " values depending on nucleation frequencies assumed in threshold or isothermal experiments may be too small by much larger factors.Dr. N. H. Pletcher (University of New England, N.S. W.) (communicated) : This comment is concerned with nucleation by foreign particles in a supersaturated environment. It may be o€ interest to note that Twomey,l working in Australia. has provided an experimental verification of the Volmer eqn. (12) and (13) relating to the nucleation efficiency of a flat surface with contact angle 8. Twomey was able to produce controlled water-vapour supersaturations ranging from 0.3 % to 100 % using a diffusion chamber containing water and dilute HC1, and, by examining the glass walls of the chamber, to detect the onset of con- densation upon them.The glass was then coated with a variety of materials to produce contact angles ranging from 15" to 80°, these contact angles being measured by ordinary methods. Throughout the whole range of contact angles examined the nucleation threshold was found to agree very well with that predicted by the Volmer theory, Eqn. (12) and (13) have also recently been generalized to give a detailed treat- ment of nucleation by small spherical particles.2 As suggested by the analysis of Reiss, referred to in Dunning's paper, the effect of particle size becomes ap- preciable when the particles are not much bigger than critical embryos. In practice, this occurs for particle diameters of the order of 0 . 1 micron and so is quite important in discussions of the nucleation activity of many aerosol systems.Dr. H. Wilman (Imperial College, London) said: With respect to the last section of Dr. Mason's paper, there is surely no reason to doubt that water molec- ules can and do condense to form solid crystalline ice directly on a substrate at a temperature below the melting point, in just the same sort of way as molecules of other materials, such as NaCl or CdI2, do. This can be seen from some of Dr. Mason's beautiful photographs of ice crystals growing on faces of mica, silver iodide, and other crystals. The hexagonal plate-like shape of the ice crystals is direct evidence that the water molecules (at the substrate temperature con- cerned) migrate considerable distances from their initial point of impact on arrival, until they are preferentially held in the deeper potential troughs at the edges of the growing crystal face, thereby prolonging the main facb further.In the slide showing ice condensing as hexagonal plates at the edges of the spiral step present on a silver iodide face, this preferential location of the ice crystals indeed corresponds well to such an interpretation. It appears that the water molecules have migrated to the edges of the silver iodide face and have been held there preferentially at such sites, just as the AgI molecules had evidently done during the formation of the AgI substrate previously; and their preferential aggregation to form ice crystal nuclei has then occurred correspondingly in these regions which are more densely populated with water molecules, at the edges of the steps on the AgI.It is thus clear that although the individual water molecules had considerable mobility at first, after their arrival on the substrate, there is no evidence of con- densation as a continuous liquid layer. Dr. N. H. Fletcher (University of New England, N.S. W.) (communicated): This comment is concerned with the nucleation of droplets from supersaturated water vapour by small ions. Eqn. (3) of Mason's paper gives the equilibrium supersaturation over a small droplet of radius r and charge q, under the assumption that the dielectric constant of the liquid is infinite, or equivalently that the liquid is sufficiently conducting that the ionic charge may be regarded as uniformly distributed over the droplet surface. This assumption may be questioned, and 1 Twomey, J.Chem. Physics, 1959, 30, 941. 2 Fletcher, J. Chem. Physics, 1958, 29, 572; 1959,31, 1136.62 GENERAL DISCUSSION it seems probable that a better model would treat the ionic charge as localized at the centre of the droplet, the material of the droplet then being regarded as behaving as an ideal dielectric whose permittivity would be very much less than the low field value for water because of saturation effects in the strong field of the ion. A more serious objection may be raised, however, to the use of this equilibrium equation to calculate nucleation thresholds. Such a procedure neglects the statistical fluctuations which are the whole basis of nucleation theory. Such fluctuations are of supreme importance in cases such as this in which the total number of molecules in a critical embryo is only about 30.The agreement of the critical supersaturation calculated in this way with the experimental value is only fortuitous and in fact the critical value of S given by (3) approaches infinity as the charge q approaches zero, instead of tending to a value near 6 as required for homogeneous nucleation. A theory of nucleation upon ions which takes proper account of statistical fluctuations and which makes use of a more realistic model for the charged droplet can be worked out along the same lines as the treatment of homogeneous nucleation leading to eqn. (2). Such a treatment was, in I'act, given as long ago as 1938 by Tohmfor and Volmer.1 The chief uncertainty in their treatment is the appropriate value to be used for the dielectric constant of the drop material.Values in the vicinity of 3 were found to give good agreement with experiment for several liquids, and such values seem reasonable in view of saturation effects. A further comment concerns the difference in nucleation eificiency between positive and negative ions. As suggested in the paper, this result is probably due to the existence of an oriented layer at a normal water surface, and in fact It is easy to show that the effect to be expected is of about the order observed. For a droplet about 10-7 cm in radius, such as those with which we are dealing, the energy of a water dipole oriented against the direction of the ionic field is -10 kT so that essentially all dipoles will be oriented parallel to the local field determined by the sign of the ion.Now there is good evidence that in a normal water surface the outermost molecular layer, and probably several beneath it, are oriented with the hydrogen ions directed inwards to the liquid. The entropy deficit associated with such an orientation restriction is about k log 2 per molecule, so that the enthalpy re- duction per molecule produced by such orientation must be more than RT log 2. A water surface in which the dipoles are pointed in the " wrong " direction should thus have its surface free energy increased by about 2 IcT log 2 per molecule or about 50 erg/cm2 to a value near 130 erglcm2. Now for a negative ion the normal surface orientation is reinforced and the Tohmfor-Volmer theory predicts a nucleation threshold at about S = 4.For a positive ion, however, the surface orientation is reversed so that the surface free energy is increased from about 80 to about 130 erg/cm2. The effect of such an increase would be to raise the value of S for the nucleation threshold by about a factor three. The experimental difference between S for ions of different sign is about a factor 1.5, which is less than the factor estimated above, but in viev of the order-of-magnitude nature of the argument used, this agreement is fairly satisfactory. A really good theory must necessarily, as pointed out in the paper, take proper account of the pseudo-crystalline nature of the growing embryo, Dr. B. J. Mason (Imperial College, London) (communicated) : I agxm, of course, with Dr. Fletcher that my eqn.(3) does not describe the rate of droplet formation on ions and that this may be discussed on a statistical basis similar to that described in the section of my paper dealing with homogeneous nucleation. I have serious misgivings about the Volmer theory of homogeneous nucleation and also its ex- tension by Tohmfor and Volmer to describe condensation on ions because i t s 1 Tohmfor and Volnier, Ann. Physik, 1938,33, 109.GENERAL DISCUSSION 63 most serious limitations are just those of eqn. (3) which €oms an essential part of the theory. My intention was therefore to discuss the limitations of eqn. (3) which is incapable of accounting €or the well-established fact o€ sign preference. I agree entirely that it is fortuitous that the supersaturation required for the formation of nuclei of critical size predicted by eqn.(3) is dose to that observed €or a detectable rate of droplet formation on -ve (but not +ve) ions and 1 am not sure that we are much better off by modifying eqn. (3) to take into account the polarization of the droplet surface by an ion in the manner of Tohmfor and Volmer and writing where 81, 82 are the dielectric constants of the surroundings and the droplet respect ivd y. What values shall we choose for 82 ? If we use the bulk values, 81 = 1 , EZ = 100, the correction is negligible, The alternative is to leave 82 as a further disposable parameter in the equation for the nucleation rate giving four unknowns, 82, ~ L V (or doLV/dr), r, (the effective initial radius o€ the ion) and S.One may assume values for EZ, CTLV and r, and use the experimental data on nucleation rates to calculate a corresponding theoretical value of S to be compared with the measured value ; or, alternatively, follow Volmer and use the bulk value of BLV, the experi- mental value of S, guess at r,, and compute the effective value of 82 = 3 which Fletcher mentions. But here we are really only camouflaging our ignorance and we must not delude ourselves that the experiments provide an independent check of the theory which is manifestly inadequate because it does not account for sign preference. My general feeling is that a disproportionate effort in nucleation physics has been spent titivating the Volmer-type theories and to manipulating the parameters to obtain agreement with selected results from rather doubtful experiments and muchtoolittle attention to designing good experiments which may ascertain the facts. Dr- W.J. Dunning (Bristol Uaiversity) said: Dr. Fletcher suggests that the basal (8001) planes of AgI and PbIz should not nucleate ice efficiently, despite their close structural resemblance to it. I should like to ask Dr. Fletcher if there is any evidence that the (0001) habit faces are 0001 faces on a molecular scale? If a b FIG. 1 .-(a) nominal ( O W ) surface ; (b) scheme of narrow corrugations simulating nominal (001) stdam. they were, consisting of ions all of the same sign, we should expect them to have very high surface energies. It seems possible that such nominal 0001 faces may consist of a number of short or narrow surfaces all of lower surface energy.For example the nominal (0001) surface may really consist of narrow corrugations formed from (1011) faces. Dr. Fletcher mentions the possibility that the barrier €or heterogeneous nucle- ation may be lowered by the effects of the vapour molecules adsorbed on the sub- strate surface. If the adsorbed film on the surface builds up in thickness, its stability will depend upon the sign of dp2ldr2, where p2 and r2 are the chemical potential and surface excess of the adsorbate. Should a fluctuation occur in which64 GENERAL DISCUSSION the film becomes thinner in one part and thicker in another, the fluctuation will be transient if dp2lpr2 is positive. On the other hand if, within a certain range of r2 values, dp2/dlT2 is negative, then the thick part will become thicker and the thin part thinner.When such surface phase separation occurs, it simplifies matters if a condition of equilibrium between the two surface phases can be introduced. A simple case would be when one of the surface phases is a liquid film and the equilibrium can be expressed by Young’s equation. In surface nucleation the question arises whether the process is such that nucle- ation occurs before the surface film is built up or whether the surface film builds up first and then collapses. At low supersaturations it might be expected that the film builds up and then collapses, whilst at high supersaturations it might be expected that nucleation occurs directly from the vapour. Dr. B. J. Mason (Imperial College, London) said: Because we now have abundant evidence that ice nucleation (and the photolytic decomposition of silver iodide) occurs preferentially at special sites on the substrate, e.g., at dislocations, edges of growth steps, re-entrant corners, etc., I doubt whether the nucleating ability of a surface can be discussed realistically in teilns of macroscopic concepts of surface energies and contact angle.We are not concerned so much with the average force field over large, flat areas of a perfect crystal face as with the much more complicated force field in the neighbourhood of imperfections. In these regions the probability of nucleation will be determined by a combination of factors : the degree to which the local geometry (e.g. a step) enables the nucleus to surmount the energy barrier for two-dimensional nucleation; the degree of misfit between the host and guest lattices; and the associated degree of strain which will determine the stability of the deposit.At supersaturations of about 10 %, our experiments show that, on large crystals of Agl, PbI2, CdIz, epitaxial deposits of ice crystals appear only on steps, etch pits, etc., and that supersaturations of about 100 % are required for nucleation on flat, perfect areas of the crystal, Silver iodide particles of radius 0.1-1 p act as ice nuclei if the air is sub-saturated relative to liquid water provided it is supersaturated relative to ice by at least 12 %. On the other hand, much smaller particles, of radius about 0.01 p, form ice crystals only if the relative humidity exceeds about 120 % which is understandable if the particles are involved in condensation followed by freezing.This is contrary to Dr. Fletcher’s deduction that such silver iodide particles must act as sublimation nuclei in natural clouds. I agree that relative humidities of 120 % are not achieved in clouds, and so the inference is that very small silver iodide particles can act only by being captured by super- cooled cloud droplets. can, however, act as sub- limation guclei. Mr. W. R. Lane (War Dept., Porton Down) (communicated) : Dr. Fletcher finds fair agreement between his calculated curve for the activity of a model AgI aerosol and activity curves measured on the smoke produced from practical generators and he quotes this in support of the validity of his approximate theory of the particle size effect.It is as well to keep in mind the fact that the activity of silver iodide smoke appears to depend upon the manner in which the nuclei are produced. For example, in experiments on the nucleation of supercooled water clouds by smoke produced (i) by vaporization of an acetone solution of silver iodide in a paraffin- burning generator of a type which has been employed in large-scale cloud seeding operations and (ii) from a small electrically-heated laboratory vaporizer, we obtained the following comparative results for smokes tested 3 min after generation. Larger particles, with r>@1 temp. of supercooled cloud O C number of ice crystals per g Ag I paraffin burner small vaporizer - 10 1.45 X 1012 2.6 X 1011 - 15 PO x 1013 6.3 x 1013 - 20 1-65 x 1013 20 x 1015GENERAL DISCUSSION 65 (If it is desired to allow for particle losses from coagulation, diffusion and sedimentation during the 3-min storage before test, these numbers should be increased by approximately 25 %.) At temperatures below about - 12°C the AgI aerosol generated from the small vaporizer shows a considerably higher activity than that from the paraffin burner, but this is not maintained at higher temperatures.In attempting to discover the reason for the different nucleation activities of AgI aerosols produced in these two ways, representative samples of each smoke were examined by electron microscopy. The smoke from the paraffin burner showed not only a lower proportion of very small individual particles than that from the laboratory device, but also numerous " stains " whose appearance sug- gested that many AgI particles had been engulfed in droplets formed, presumably, in the burning of the acetone solution in the generator.The lower activity of the smoke from the paraffin burner may be associated with these particle character- istics. Dr. E. R. Buckle (Imperial College, London) said: Since the completion of the work on alkali halides reported in my paper, we have been studying halides of heavier metals in the cloud chamber. Freezing thresholds for AgCl, CdI2, PbC12 and PbI2 have been located by observing the onset of twinkling and by rnicro- scopical examination of cloud fall-out. Crystalline particles in PbI2 clouds show very marked twinkling when first seen, but on continued circulation further growth takes place and the twinkling becomes less distinct.This appears to be due to dendritic growths formed by direct sublimation on the hexagonal prisms which crystallize from the melt. The resulting crystals closely resemble those common in snowflakes. Dr. B. J. Mason (Imperial College, London) said: The appearance of twinkling particles is widely accepted as a reliable indication of the onset of crystallization in clouds of supercooled droplets. But in order to twinkle, the particles must develop flat crystal faces o€ diameter several times the wavelength of the incident light, i.e. of at least a few microns. The time taken for the appearance of such faces will depend upon the temperature and supersaturation of the vapour.When watching micron-size water droplets falling through a temperature gradient in air I have noticed that their freezing at - 40°C is indicated by a sudden brightening of the droplet and after a delay of a second or two they become transformed into twinkling ice crystals. We have also observed that thin circular disc-like ice crystals deposited from the vapour or grown from the melt develop into six-sided plates more readily when they are growing slowly. I imagine a frozen droplet to be surrounded by an almost infinite number of small crystal faces of different orientation. As growth proceeds, the high-index faces grow faster and tend to grow out until, ultimately, the particle is surrounded by a small number of low-index faces and we have a recognizable crystal.If, however, the particle continues to grow rapidly, imperfections are built in and lead to the constant regeneration of high-index faces, and the appearance of fiat faces large enough to cause specular reflection is delayed. This also may be true at low temperatures when the surface migration of molecules is much reduced and the condensation may, in extreme cases, acquire an amorphous structure. I should like to ask Dr. Buckle if he recognizes this as a problem in determining the onset of crystallization in his high-temperature smokes. Also, does he believe that the crystals which he observed always arose from the freezing of supercooled droplets or was there any evidence to suggest that some may have formed by direct sublimation from the vapour phase? Dr.E. R. Buckle (Imperial CoZZege, London) said: In reply to Dr. Mason, one of the most important factors determining the rate of growth of crystals is the dissipation of the heat of crystallization. I think that this can occur very rapidly in our experiments on alkali halides, since it is calculated 1 that the hot particles 1 Buckle and Ubbelohde, to be published. C66 GENERAL DISCUSSION generated near the supersaturator cool through 300-400°C in periods of the order of a millisecond. It is likely that linear crystallization velocities in molten alkali halides are sufficiently high to cause the melt droplets to solidify completely in a very small fraction of the time of observation of clouds. Regarding the stage at which crystal faces suitable for specular reflection appear, I agree that this may occur quite late in the growth process, but in view of the rapidity of freezing as a whole this would not affect the measurement of Ts.As mentioned in my paper, at low temperatures the crystals may be formed by sublimation. There was fairly conclusive evidence for this in some cases, since it was occasionally possible to obtain clouds at low temperatures without melting the supersaturator bead. By examining fall-out at furnace temperatures around T,, however, we have established that cloud particles are invariably molten just above this temperature and crystalline just below it. Moreover, as was also stated in the paper, theoretical calculations, based on the Becker-Doering theory and reasonable estimates of vapour supersaturation, show that at temperatures just above the melting point of the salt nucleation of droplets should occur very rapidly.Since the supersaturated vapour is initially at even higher temperatures, droplets evidently provide the route by which crystals are formed at T,. Dr. €3. Wilmwn (Imperial College, London) said: I would like to emphasize that, as our own results illustrate, a spherical particle shape does not necessarily by itself mean that the particles have condensed as liquid. For example, in our smoke deposits from arcs between metal electrodes in air, the practically spherical shape of the Ta2Q5 particles is surely most unlikely to be due to condensation of either Ta or Ta2O5 as molten particles, since both these materials have very high melting points (even the crystallization temperature of initially amorphous Ta205 is as high as 700°C). From many electron diffraction investigations we know that if a vapour con- denses to form a solid deposit on a substrate (or for aerosols, the initial nuclei) at a low enough temperature, the deposit is amorphous ; but in a higher substrate- temperature range the deposit is formed as small crystals (e.g., -5OA diam.), but corresponding to the still only very low degree of mobility of the atoms or molecules on the deposit surface, there is then practically no facet development on the crystals, so that with the aerosols the particles must develop roughly spherical shape, In a higher temperature range again, the higher mobility of the atoms or molecules after arrival on the deposit leads to formation of much larger crystals and these develop plane faces which are extended by the atoms or molecules migrating after arrival until preferentially held at the edges of surface sheets of close-packed atoms.Above the melting-point, finally, the deposit is condensed directly in the liquid state, again as spheres; and when such spheres subsequently cool, they may retain this shape if the cooling is rapid enough and extensive enough to reduce the atomic motion to a level at which crystallization, or recrystallization, cannot occur to any appreciable extent. However, in the particular case of condensation of PbI2 from the vapour, to which Dr. Buckle has just referred concerning his cinematograph film, it seems clear that the temperature at which the PbI2 was condensed must have been higher than the melting point, so that the spherical particles must have condensed indeed as liquid. This is similar to our own observations in the condensation of Pb and Bi. On the other hand, when crystalline plates are observed to be formed, this is usually evidence that either all or most of the deposition has occurred directly from vapour to crystalline solid, unless initially liquid drops are formed and cool relatively slowly so that facets develop by, in effect, a recrystallization type of rearrangement of the atoms or molecules into a configuration of lower potential energy than that of the spherical initial shape. Dr. E. 33. Buckle (Imperial College, London) (communicated) : Incidentally, it is possible that twinkling may not always be simply it reflection phenomenon.GENERAL DISCUSSION 67 We have observed 1 an effect in dense clouds, both of liquid and of solid particles, due to the temporary eclipsing of one particle by another. The effect is visible in the telescope as a momentary brightening of the nearer particle. In turbulent clouds, or in clouds composed of sub-micron particles undergoing visible Brownian translations, this momentary brightening occurs repeatedly, and closely resembles the twinkling of isolated crystals. Another reason for suspecting that twinkling is not always due to specular reflection is that we have also observed sub-micron particles to twinkle, even when completely isolated. 1 Buckle and Ubbelohde, I. U.P.A. C. Symp. Thermodynamics (Wattens, 1959).