Central nervous system (CNS)-directed prophylactic intrathecal (IT) therapy is indicated in patients with Burkitt and acute lymphoblastic lymphoma. Its role in diffuse large B cell lymphoma (DLBCL), a heterogeneous subtype, is less well defined. While addition of rituximab to standard cyclophosphamide-hydroxydaunorubicin-oncovin-prednisone (CHOP) chemotherapy (R-CHOP) has improved the outcomes of DLBCL patients, its role in reducing CNS relapse is unclear. We aim to (1) evaluate the clinical risk factors predictive of CNS relapse, (2) the role of rituximab in influencing CNS relapse, and (3) role of intrathecal prophylaxis. Four hundred ninety-nine patients with DLBCL from 2000 to 2008 were included (CHOP 179 vs. R-CHOP 320). IT prophylaxis was administered to 82 patients based on our institution's guidelines. Baseline characteristics between CHOP- and R-CHOP-treated patients were similar. Although R-CHOP significantly increased the complete remission rate from 71% to 81% (P < 0.01), CNS relapse rates remained unchanged (R-CHOP 6% vs. CHOP 5.1%). On multivariate analysis, poor performance status (Eastern Cooperative Oncology Group >1; hazard ratio (HR) = 2.01, 95% confidence interval (CI) 1.29-3.14), failure to attain remission (non-complete response (CR) vs. CR: HR = 2.39, 95% CI = 1.03 to 5.51), testicular (HR = 6.67, 95% CI = 1.62 to 27.53), kidney (HR = 20.14, 95% CI = 5.23 to 77.46), and breast involvement (HR = 6.14, 95% CI = 1.61 to 23.37) were each independently predictive of CNS relapse. Use of IT prophylaxis did not appear to decrease CNS relapse. Median survival after CNS relapse was 3.2 months. CNS relapse, a fatal event, remains a challenge in R-CHOP-treated patients. IT prophylaxis may not be sufficient to reduce CNS relapse, and strategies including systemic agents with high CNS penetration should be evaluated in high-risk patients identified in this study.