The problem of how to estimate probabilities has interested philosophers, statisticians, actuaries, and mathematicians for a long time. It is currently of interest for automatic recognition, medical diagnosis, and artificial intelligence in general. The main purpose of this monograph is to review existing methods, especially those that are new or have not been written up in a connected manner. The need for nontrivial theory arises because our samples are usually too small for us to rely exclusively on the frequency definition of probability. Most of the techniques described in this book depend on a modern Bayesian approach. The maximum-entropy principle, also relevant to this discussion, is used in the last chapter. It is hoped that the book will stimulate further work in a field whose importance will increasingly be recognized. Methods for estimating probabilities are related to another part of statistics, namely, significance testing, and examples of this relationship are also presen