Background: The Geriatric Nutritional Risk Index (GNRI) is a nutritional screening method primarily developed for elderly people; it is also reported to be useful for predicting mortality in patients on maintenance dialysis. However, it is unclear whether it is useful at the time of dialysis initiation, which is accompanied by large weight fluctuations and unstable nutritional status. Methods: The study included 1524 patients with chronic kidney disease who commenced dialysis therapy at 17 centers. Patients commenced dialysis between October 2011 and September 2013 and were followed up until March 2015. Results: We analyzed 1489 patients whose GNRI could be calculated and whose prognosis was clear. The mean GNRI was 87.60 (median 87.86). We divided patients based on the median value into a high (H) and low (L) group. The H group included 728 patients (mean GNRI 95.2 ± 4.9, mean age 65.8 ± 13.2 years, 69.3% men), and the L group included 761 patients (mean GNRI 80.3 ± 6.1, mean age 69.1 ± 12.8 years, 66.0% men). Mortality was significantly higher in the L group (L, 22.2% vs. H, 12.6%, P < 0.001). The rates of infection-associated death in the L group was significantly higher (L, 5.5% vs. H, 1.9%, P < 0.001), although no significant difference was observed regarding cardiovascular disease-associated death (L, 7.6% vs. H, 5.2%, P = 0.059) and malignancy-associated death (L, 3.0% vs. H, 3.0%, P = 1.000). Multivariate analysis showed an association between GNRI and all-cause mortality (HR 0.9852, 95%CI 0.9707-0.9999, P = 0.049) and infection-associated death (HR 0.9484, 95%CI 0.9191-0.9786, P < 0.001). Conclusions: GNRI is useful for predicting mortality even at the time of dialysis initiation. Among the causes of death, GNRI was strongly associated with infection-associated death.