A speech recognition method and apparatus is disclosed in which outputs a confidence score indicative of the posterior probability of an utterance being correctly matched to a word model. The confidence score for the matching of an utterance to a word model is determined directly from the generated values indicative of the goodness of match between the utterance and stored word models utilizing the following equation:.function..sigma..times..function..times..function..sigma..times- ..function. ##EQU00001## where S(x|w) is the match score for the correlation between a signal x and word w and .sigma. is an experimentally determined constant.