This paper illustrates how the duality theory of geometric programming and the special methods available for solution can be used to obtain maximum likelihood estimates for multinomial probabilities.