Background The association between N-terminal pro brain natriuretic peptide (NT-proBNP) level and long-term mortality in Japanese hemodialysis patients has not been fully assessed. Methods This prospective, multicenter study included 1428 hemodialysis outpatients. Baseline NT-proBNP levels were measured at the first hemodialysis session of the week and participants were followed for 5 years. The areas under the curve were calculated from receiver operating characteristic curves. Groups determined by quartiles of baseline NT-proBNP level were assessed by the Kaplan-Meier method and log-rank test. The association between NT-proBNP level and mortality was assessed using multivariate Cox proportional hazards models. Results During the 5-year follow-up, we observed 370 deaths and 256 censored cases. The areas under the curve of pre-hemodialysis NT-proBNP for all-cause mortality and cardiovascular disease mortality after 1 year were 0.75 and 0.78, respectively, and significantly greater than the areas under the curve at the 3- and 5-year follow-up. Cut-off values for all-cause mortality and cardiovascular disease mortality after 1 year were 4550 and 5467 ng/L, respectively (sensitivity: 82% and 81%; specificity: 59% and 64%). Kaplan-Meier survival analysis showed that the group with pre-hemodialysis NT-proBNP >= 8805 ng/L had increased all-cause mortality (P < 0.001) and cardiovascular disease mortality (P < 0.001). Finally, multivariate Cox analysis showed that NT-proBNP level was associated with all-cause mortality (P < 0.001) and cardiovascular disease mortality (P = 0.004) independently from other clinical parameters. Conclusion NT-proBNP is a useful marker to predict both all-cause and cardiovascular disease mortality in hemodialysis patients.