Gene-expression microarray is a novel technology that allows to examine tens of thousands of genes at a time. For this reason, manual observation is not feasible anymore and machine learning methods are progressing to analyze these new data. Specifically, since the number of genes is very high, feature selection methods have proven valuable to deal with this unbalanced – high dimensionality and low cardinality – datasets. Our method is composed by a discretizer, a filter and the FVQIT (Frontier Vector Quantization using Information Theory) classifier. It is employed to classify eight DNA gene-expression microarray datasets of different kinds of cancer. A comparative study with other classifiers such as Support Vector Machine (SVM), C4.5, naïve Bayes and k-Nearest Neighbor is performed. Our approach shows excellent results outperforming all other classifiers.