A statistical pattern recognition technique for classification by supervised learning was developed and applied to automated recognition of marine mammal sounds [J. Acoust. Soc. Am., 101(3), March 1997]. Training data identified by human experts are characterised by occupancy statistics associated with a multiple-resolution, binary partition of the unreduced observation space. Classification of a new sample is performed by Bayesian inference applied to these occupancy statistics. The classification algorithm is implemented in a simple, highly efficient computer program. The present work describes efficient encoding of the training data distributions for large numbers of training samples and classes, and efficient evaluation of a posteriori probabilities of class membership for classification of new samples. Data storage requirements and computational efficiency of the classification algorithm are compared with theoretical bounds.