When less than 30 observations are available it is customary to employ Student's “t” distribution to establish confidence limits within which the mean of the entire population lies with an assigned probability. However with a sample size as small as two, and even with a non conservative probability of 90 per cent., the bounds of the mean are often so wide as to be of little practical value. With the assumption, frequently met in practice, that the coefficient of variation is not more than 20 per cent., this paper gives practical bounds within which the true mean is likely to lie, when only two observations are available. The theory is applied to concrete testing and some experimental justification is given.