This short communication deals with a special kind of hash function called “Bloom filters.” These filters are used, for example, to search a differential file containing updates to a main file. The paper recalls theoretical results on the number of errors, that is, accesses to the differential file when the record searched for has not been actually modified. The author then presents some experimental results to support the claim that the choice of suitable hash functions gives a practical algorithm. This paper is easy to read and may prove useful to programmers wishing to achieve the performances of Bloom filters predicted by theory.