Microarray technology has proposed a powerful tool in dealing with the heterogeneity of disease. Currently, many methods in the field are based on traditional hierarchical clustering to discover subtypes of disease using a large number of genes on microarray.However, they did not considered that large unrelated noise (genes)may mask significant partitions and correlations of disease samples. To avoid the shortcoming, this paper presented a heterogeneous analysis based on coupled two-way clustering (HCTWC) to search interesting gene signature and find the natural partitions of disease samples. The method was applied to diffuse large B-cell lymphoma (DLBCL) microarray dataset. By identifying significant gene signature, we were able to discover the two new subtypes of DLBCL with survival rate 55% and 25% respectively. The results showed that HCTWC had the potential to be a powerful tool for solving the heterogeneity of disease on gene expression profile.