Microarray gene expression techniques have recently made it possible to offer phenotype classification of many diseases. One problem in this analysis is that each sample is represented by quite a large number of genes, and many of them are insignificant or redundant to clarify the disease problem. The previous work has shown that selecting informative genes from microarray data can improve the accuracy of classification. Clustering methods have been successfully applied to group similar genes and select informative genes from them to avoid redundancy and extract biological information from them. A problem with these approaches is that the number of clusters must be given and it is time-consuming to try all possible numbers for clusters. In this paper, a heuristic, called K-means+, is used to address the number of clusters dependency and degeneracy problems. The result of our experiments shows that K-means+ method can automatically partition genes into a reasonable number of clusters and then the informative genes are selected from clusters.