This article presents a procedure and a table for selecting sample size for simultaneously estimating the parameters of a multinomial distribution.The results are obtained by examining the "worst" possible value of a multinomial parameter vector, analogous to the case in which a binomial parameter equals one-half.