On the problem of two samples