Diffuse large B-cell lymphoma (DLBCL) has been classified into different prognostic subgroups using immunohistochemistry in Western populations. However, the applicability in Chinese patients of these subgroups was unclear. We collected 116 specimens and performed immunohistochemical staining for CD10, BCL-6, MUM1, CD138, and CD5, and the results were classified into subgroups according to 3 different algorithms. We then analyzed the subgroups' correlation to patient survival. Expression of CD10 and BCL-6 predicted favorable 5-year OS (70 and 62.5%, respectively) and PFS (64.3 and 61.5%, respectively) rates. In contrast, the expression of MUM1 predicted unfavorable 5-year OS (23.1%) and PFS (17.9%) rates and was also independent of other markers. All algorithms led to useful subclassifications. Using Hans' algorithm based on CD10, BCL-6, and MUM1, the non-germinal center (GC) subgroup (66.4%) had worse 5-year OS (29.8%) and PFS (26.7%) rates than did the GC subgroup. Likewise, using Muris' algorithm based on CD10 and MUM1, fewer non-GC cases (27%) showed poorer OS (20.3%) and PFS (16.2%) rates than did GC cases, an effect that was independent of both the International Prognostic Index, a clinical indicator, and treatment. It identified a subgroup with a high-risk of death and seemed to be applicable in our series. In conclusion, these algorithms can be used effectively in Chinese patients with DLBCL.