We apply the decision tree algorithm C4.5 to gene expression data in order to induce decision trees for identification of breast cancer patients. Using expression data from 108 known breast cancer-related genes for 75 patients with various diseases of the breast, we are able to induce decision trees with 89% accuracy in separating cancer from non-cancer patients in a cross-validation test. We also show that by inducing a separate decision tree for each cancer-related gene, and using the expression level of the individual gene as the decision variable, it is possible to obtain decision trees which aid the understanding of signaling pathways involved in breast cancer. In addition, we also show that the C4.5 algorithm is able to identify key breast cancer genes when decision trees are induced on expression data sets containing randomly selected genes. This result indicates that it is possible to make biological discoveries when applying decision tree algorithms to large sets of gene expression data in diseases where the genetic basis is not well characterised.