Objective: In this subanalysis of a phase III trial using three categorized doses of S-1, the influence of the actual doses on safety and efficacy was evaluated. Methods: We compared the efficacy and safety of the S-1 or gemcitabine plus S-1 combination (GS) arm between the top 10% group and the bottom 10% group according to the initial doses of S-1: ≥77.6 versus ≤65.9 mg/m2/day (n = 28 vs. 28) in the S-1 arm, and ≥65.1 versus ≤53.8 mg/m2/day (n = 27 vs. 28) in the GS arm. Results: Overall and progression-free survival were not significantly different between these two groups: hazard ratios of 0.818 and 0.761 with p values of 0.498 and 0.330 in the S-1 arm, and hazard ratios of 0.836 and 0.759 with p values of 0.557 and 0.323 in the GS arm, respectively. Incidences of grade 3-4 hematological toxicities were significantly higher in the top 10% group than in the bottom 10% group: 42.9 versus 14.3 and 85.2 versus 57.1%, with p values of 0.037 and 0.037 in the S-1 and the GS combination arm, respectively. Conclusions: Higher actual doses of S-1 were associated with a higher incidence of hematological toxicity even in the same dose setting.