Distributional learning enables listeners to form phonetic categories by extracting statistical regularities from speech input. Younger Cantonese speakers can acquire the Mandarin level-falling (T1–T4) contrast through distributional learning, with bimodal exposure facilitating category formation and unimodal exposure suppressing it, and with fine-grained pitch sensitivity predicting success. However, aging is associated with declines in pitch sensitivity and phonetic boundary formation, which may disrupt this process. This study examined whether Cantonese-speaking older adults exhibit distributional learning of Mandarin T1–T4 and whether individual cognitive factors predict learning success. Sixty-four participants completed a pretest–training–posttest procedure with bimodal or unimodal exposure. While older adults improved in tone discrimination, no group differences emerged. Further analysis showed that those with lower pitch-related auditory memory failed to learn from unimodal input. On the other hand, fine-grained pitch perception abilities did not predict learning outcomes. These results suggest that older adults may rely on alternative learning mechanisms, such as memory-based strategies, when exposed to ambiguous input distributions. The findings indicate a shift from perceptual encoding to memory-driven processing in aging and highlight the limits of passive statistical learning in older adulthood.