Microarray data analysis is notorious for involving a huge number of genes compared to a relatively small number of samples. Detecting the most significantly differentially expressed genes under different conditions, or gene selection, has been a central focus for researchers. The gene selection problem becomes more difficult when the numbers of samples under different conditions vary significantly, or are unbalanced. A novel model-free and stable gene selection method is proposed in this paper, i.e., the method does not assume any statistical model on the gene expression data and it is not affected by the unbalanced samples. The method has been evaluated on two publicly available datasets, the leukemia dataset and the small round blue cell tumor dataset, where the experimental results showed that the proposed method is efficient and robust in identifying differentially expressed genes.