We present a supervised regression model for classifying complementary DNA microarray measurements and a method to rank genes according to their importance for the classification. To allow for a supervised regression model with no overfitting, we reduce the dimensionality of the measured samples (as given by number of genes) by principal component analysis and use the ten dominant components as inputs to multilayered perceptron models. We classify samples using a threefold validation procedure, and the procedure is repeated such that numerous multilayered perceptron models are calibrated. Thus each sample is in many validation sets and all these predictions are used as a committee to classify a sample. The sensitivity of the classification on the different genes is determined by the absolute value of the partial derivative of the output with respect to the gene expressions, averaged over samples and multilayered perceptron models. In this way the genes can be ranked. The method is successfully applied to classify solid tumors in the category of small blue round-cell tumors as well as sporadic breast tumors. In addition, the method correctly classifies additional blind tests and also identifies whether a blind test is different from the disease categories used for calibration.