An upperbound on the probability of error in classifying a pattern using Bayesian decision criterion is obtained in terms of its divergence and it is shown that the maximization of the divergence of the pattern minimizes this upperbound. Furthermore, a relationship between the divergence of a pattern and its nearest neighbor classification risk is presented.