Diffuse large B-cell lymphoma (DLBCL) is a genetically heterogeneous disease with complex tumor microenvironment (TME) alterations. However, immune cell signatures of TME and their prognostic value remain unclear in DLBCL. We aimed to identify high-risk DLBCL with specific immune cell signatures in TME. Clinical and gene expression data of DLBCL patients were obtained from previously reported retrospective datasets in Gene Expression Omnibus (GSE10846 and GSE53786 ) and a multi-center prospective clinical trial NHL001 (NCT01852435). Patients treated with cyclophosphamide, doxorubicin, vincristine, and prednisone (CHOP) regimen (n = 159) from GSE10846 were referred as training cohort for CHOP regimen, while patients treated with rituximab-CHOP (R-CHOP) regimen (n = 192) from GSE10846 as training cohort for R-CHOP regimen. Patients from NHL001 (n = 68) and GSE53786 (n = 57) were referred as validation cohorts for R-CHOP regimen. CIBERSORT was applied to estimate the relative proportions of 22 subtype of immune cells. We established a prognostic model for model for R-CHOP regimen included Age, performance status, lactate dehydrogenase, T cells follicular helper and macrophages M0, defining a low-risk group with 2-years OS of 92.9% and a high-risk group with 2-years OS of 52.5% (HR 6.57 [3.27-13.18], p < 0.0001). Immune cell signatures could be used as prognostic markers and provided further insights for individualized immunotherapeutic strategies in DLBCL.