The argument which is developed here starts from the computation of the probability that a word will be absent from an exhaustive random sample drawn from a corpus whose complete frequency distribution is known. This probability is the basis of the formula put forward, more than 20 years ago, by C. Muller. Muller's formula is compared here to its equivalent in the hypergeometric model. Two studies were carried out: first the computation of vocabulary increase in corpuses and, secondly, the comparison between Muller's values and averages obtained by drawing a large number of random samples from several corpuses. It is thus demonstrated that this formula is a good approximation of the hypergeometric law. The need for associating standard deviations to the computed values is also emphasised since confidence levels have to be taken into account.
