Fluid-flow properties of porous media, such as permeabilitykand irreducible water saturationSwi,can be estimated from water1Hnuclear magnetic resonance (NMR) relaxation data, but there are basic questions regarding data processing and interpretation. We found thatSwiandkare better estimated if different forms of “average” relaxation time are used. NMR longitudinal relaxation data for a suite of 106 water-saturated clean sandstones were used. Sandstones represent a specialized class of porous media, where even for small porosity, substantially all pore space is connected. The sandstones exhibit distributions of relaxation times ranging over factors from at least 10 to more than103.We tried several forms of “average” relaxation timeT.One family ofTs is⟨Tp⟩1/p,wherelim p→0gives the geometric mean. The best estimator we found forSwiuses a form of average relaxation timeonly, rather than relaxation time cutoff. The time used can be any of several forms ofT,giving more emphasis toshort timesthan the geometric mean does. On the contrary, the bestTfor estimating permeability without other information is precisely the geometric mean. The best estimates of permeability came from fits ofln (k/&fgr;)usingTs with emphasis atslightly longer times. WhileSwiis better estimated by using all the data points (starting from our minimum 0.4 ms),kis better estimated by starting at a few ms, that is by ignoring anon-negligiblefraction of the signal for some samples. These results can be obtained also by using computations that do not need to invert multiexponential relaxation data, and good results are obtained even with only a few data points. The results are compatible with the reasonable picture, where high surface-to-volume pores, giving signal components with short relaxation times but not contributing to the permeability, are important in determining the fraction of the wetting phase which remains trapped in the solid matrix after displacement with a nonwetting phase. ©1997 American Institute of Physics.