This paper proposes a new feature selection method, called weighted punishment on overlap (WEPO), for microarray data analysis. WEPO takes advantage of parametric and nonparametric estimations to rank genes sensitively despite the limited number of samples. The proposed method was implemented and applied to three datasets. Based on informative testing, sensitivity testing, and significance testing, we analyzed the performance of WEPO and five well-known feature selection methods. Analysis results indicate that genes selected using WEPO are more informative and sensitive than those selected using the other surveyed methods and are statistically significant. Biological results also show that WEPO is able to identify meaningful genes for test data sets. The analysis and experimental results indicate that WEPO is a promising approach to select important genes in a microarray data.