A two-step procedure for comparing two binomial probabilities is presented. The method is compared with the exact test of FISHER. In average it needs essentially smaller sample sizes. I n situations like drug screening it is desired to minimize the type I1 error rates for great differences of the binomial probabilities and to maximize it for small differences, simultaneously. In this point the method is superior to'the FISHER test. Examplea from drug screening demonstrate the advantages of the method.