We examined how accent variability during training on minimal-tone-contrast Mandarin words affects English listeners' subsequent generalization of tone categorization and discrimination to new talkers and accents. English listeners underwent 6 days of training on 16 pseudowords (4 tones 4 sets) produced by 12 talkers of either Beijing (n = 24) or a mix of Beijing, Yantai, and Guangzhou (n = 24) accents. In a post-training test, they used tone contour icons to categorize the tones produced by new talkers with a familiar and an unfamiliar accent, and to discriminate all possible tone contrasts for the new talkers. While both training groups discriminated all six tone contrasts for both new talkers, only the multiple accent group reliably categorized all four tones. The single accent group failed to correctly categorize the falling tone in both generalization tests. These results suggest that accent variability during tone-word training can facilitate subsequent tone categorization.