This is version 0.3 of the Discriminant Analysis Toolbox with major bug fixes and additional functionality. A new class hierarchy is introduced with a single parent class for all classification objects. In addition to linear, quadratic and logistic discriminant analysis, limited feed-forward neural network classification is also made possible using the SOFTMAX multinomial deviance critereon for error minimisation. Additional functions include k-fold cross-validation, confustion matrices and several plotting functions. Many of these algorithms are described in B. D. Ripley (1996) Pattern Recognition and Neural Networks.