The paper contains a proposition of a method for classifying the natural languages according to the number of their segmental phonemes. In order to solve this problem the attempt is made to find the distribution of the random variable ‘number of segmental phonemes in the inventory’ for 600 languages. The hypothesis by Sigurd, according to which this distribution is a normal one, cannot be accepted. The author comes to the result that the distribution of the number of phonemes is a gamma distribution with the parameters p = 4 and b = 1.05. This result has to undergo further examination. After the form of the distribution has been established, a confidence interval is constructed which fixes the limits of the three intended classes. The author also discusses the possible linguistic-systematic value of the proposed classificat