Feature subset selection (FSS) methods play an important role for cancer classification using microarray gene expression data. In this scenario, it is extremely important to select genes by taking into account the possible interactions with other gene subsets. This paper shows that, by accumulating the evidence in favour (or against) each gene along a search process, the obtained gene subsets may constitute better solutions, either in terms of size or in predictive accuracy, or in both, at a negligible overhead in computational cost.