We propose a variational method to describe some families of probability distributions. The given probability distributions are to be found from the knowledge of the expected values of some variables. To do that we restate the problem as a constrained linear inverse problem, which we then proceed to solve by the method of maximum entropy on the mean.