It is regularly asserted that the frequency of orders received by a stockist can be represented as a Poisson process, but very little corroborative evidence has been published. This paper presents some results supporting the assumption. An adaptation to the standard testing methodology is presented which overcomes the complications arising from the necessity of sampling from time series data. The new approach is applied to a large range of stock-keeping units. Except for the extreme tail of the distribution, which is of little interest for most inventory applications, it is found that the frequency of orders is well represented by a Poisson process.