In this paper we present a new approach for classification of microarray data. Our methodology consists of two steps: an attribute selection, which aims at selection of the most informative genes, and a classification of expression profiles, which is carried out by weighted voting, a novel instance-based classifier based on Rough Set Theory. Attribute selection consists of two stages - initial selection, where each attribute is evaluated individually, and attribute refinement, where the attributes are further reduced by means of genetic computations. The effectiveness of the proposed approach was verified on six different microarray datasets, and compared with attribute selection and classification based on nearest neighbor classifier.