Cluster analysis of gene expression data is one of the most useful tools for identifying biologically relevant groups of genes, however, gene expression data suffer severely from the problems of measurement noise, dimension curse, high redundancy between genes, and the functional annotation of genes is incomplete and imprecise. These properties lead to most of the traditional clustering algorithms are very sensitive to the initialization, and are likely to get the local result, and also made the analysis results lacking of stability, reliability and biological interpretability. In the present article, we propose incorporating the data density and gene functions into distance-based clustering method, which can get more stable and reliable results, especially in discovering gene set with completely unknown function.