The grain-size distribution of aeolian dune sands in the Thar Desert, India was analyzed and compared with three model distributions -log-normal, log-hyberbolic and log-skew-Laplace -to determine the best-fi t statistical model. In total, 51 samples were collected along a single transect over a transverse dune, of which 15 were from the stoss side, 12 from the crest and 24 from the lee side. Samples were collected during a calm period in the afternoon of a winter's day. It was observed that of these 51 samples, 33 fi t best to a log-hyperbolic distribution, 14 fi t best to a normal distribution and only four fi t best to a Laplace distribution. However, it was further observed that of 24 samples from the lee side, 13 fi t best to a normal distribution, eight fi t best to a hyperbolic distribution, and three fi t best to a Laplace distribution. Of 12 samples from the crest of the dune, 11 fi t best to the log-hyperbolic distribution, only one to the Laplace distribution but none to a normal distribution. Of 15 samples from the stoss side of the dune, only one sample best-fi ts a normal distribution, 14 fi t best to a log-hyperbolic distribution, and none best fi t to a Laplace distribution.During sample collection a calm period prevailed and there was no dusty wind. It was therefore assumed that in the initial stage a mixture of coarse, medium and fi ne sands was laid down on the stoss side of the dune. As wind speeds increased and saltation started, the coarser fractions were segregated and lagged behind on the stoss slope. In the fi nal stage when the remaining intermediate and fi ner fractions reached the dune crest, the fi ner fractions were winnowed away to suspension from the crest of the dune. As a result, a narrow range of intermediate sized sediments was deposited by rolling down the lee side to explain the development of log-normality. In such a situation, both the coarser and fi ner fractions, to which the skewed distributions can be attributed, are separated from the initial mixture of coarse, intermediate and fi ne fractions. Hence the main criteria for the development of a normal distribution is the lack of skewed fractions and the concentration of the narrow, intermediate size fractions in the fi nal grain size distribution. This is also corroborated with the index of symmetry, which is a measure of the difference between the angle of two slopes of the hyperbolic distribution as represented by the coarser and fi ner fractions.