This retrospective cohort study aimed to evaluate primary treatment and recent survival trends in patients with primary diffuse large B-cell lymphoma of central nervous system (CNS) from 1995 to 2016. Using the SEER data, patients diagnosed with non-HIV-associated primary central nervous system lymphoma (PCNSL)–diffuse large B-cell lymphoma (DLBCL) aged ⩾18 years between 1995 and 2016 were identified. The year of diagnosis was divided into the time period-1 (1995–2002), the time period-2 (2003–2012), and the time period-3 (2013–2016). Chi-square tests, the Kaplan–Meier method, log-rank test, and Cox regression model were used in the analysis. Overall, 3760 patients were included. Both the use of radiotherapy alone and the application of combined chemoradiotherapy decreased significantly, following the wider use of chemotherapy alone during 1995–2016. There was a significant improvement in PCNSL cause-specific survival (CSS) (period-1: 13 months vs. period-2: 19 months vs. period-3: 41 months, p < 0.001). Survival of patients aged above 70 years did not change from the time period-1 to the time period-2 (p = 0.101). However, there was an increase in CSS from the time period-2 to the time period-3 in the elderly patients (period-2: 5 months vs. period-3: 9 months, p < 0.001). On multivariable analyses, diagnosed in the time period-3 was significantly and independently associated with better CSS (hazard ratio 0.577, 95% confidence interval 0.506–0.659, p < 0.001). Our analysis shows the use of radiotherapy in the treatment of PCNSL has waned over the study span. There was a significant improvement in CSS during 1995–2016, which reflected developments in treatment over time. The elderly patient population also gained a significant CSS benefit in the most recent period.