Background: The interaction of programmed death-1 protein (PD-1) and programmed death-1 ligand (PD-L1) produces immunosuppressive activity, protecting tumor cells from anti-tumor immunity and possibly releasing soluble PD-L1 (sPD-L1) from PD-L1 expressing tumor cells. Therefore, we measured serum levels of sPD-L1 in patients with primary central nervous system lymphoma (PCNSL) and explored its clinical implications. Methods: Sixty-eight patients with newly diagnosed PCNSL had diffuse large B-cell lymphoma and were treated with high-dose methotrexate-containing chemotherapy. The measurement of sPD-L1 and cytokines was performed using serum samples archived at diagnosis, and the tissue expression of PD-L1 was also analyzed from archived paraffin-embedded tissue blocks. Disease relapse, progression-free survival (PFS), and overall survival (OS) were analyzed according to the extent of sPD-L1 in serum and PD-L1 in tissue. Results: The median level of serum sPD-L1 (0.429 ng/mL) was higher than in healthy control patients (0.364 ng/mL). The occurrence of relapse was more frequent in the high sPD-L1 (78%) than the low sPD-L1 group (50%), though the groups did not have different clinical or pathological characteristics at diagnosis. As a result, the OS and PFS for the high sPD-L1 group were significantly lower than those in the low group. PD-L1-positive tumor cells were found in 35 patients (67%), and the extent of PD-L1-postive tumor cells was positively associated with serum sPD-L1 levels (r = 0.299, P = 0.031). Among the 34 cytokines analyzed, only the serum level of IL-7 correlated with the serum level of sPD-L1 (r = 0.521, P < 0.001). Conclusions: Serum levels of sPD-L1 could reflect the expression of PD-L1 in PCNSL tumor cells and predict patient survival outcomes. Therefore, sPD-L1 in serum could be a feasible biomarker for determining a risk-adapted treatment strategy for PCNSL patients.