Microarrays, which allow for the measurement of thousands of gene expression levels in parallel, have created a wealth of data not previously available to biologists along with new computational challenges. Microarray studies are characterized by a low sample number and high feature space with many features irrelevant to the problem being studied. This makes feature selection a necessary pre-processing step for many analyses, particularly classification. A Genetic Algorithm and Artificial Neural Network wrapper approach is implemented to find the highest scoring set of features for an ANN classifier. Each generation relies on the performance of a set of features trained on an ANN for fitness evaluation. A publically-available leukemia microarray data set (Golub et al., 1999), consisting of 25 AML and 47 ALL Leukemia samples, each with 7129 features, is used to evaluate this approach. Results show an increased performance of selected features over the classifier from Golub et al. 1999.