Purpose: Evidence suggests that extensive experience with lexical tones or musical training provides an advantage in perceiving nonnative lexical tones. This investigation concerns whether such an advantage is evident in learning nonnative lexical tones based on the distributional structure of the input. Method: Using an established protocol, distributional learning of lexical tones was investigated with tone language (Mandarin) listeners with no musical training (Experiment 1) and nontone language (Australian English) listeners with musical training (Experiment 2). Within each experiment, participants were trained on a bimodal (2-peak) or a unimodal (single peak) distribution along a continuum spanning a Thai lexical tone minimal pair. Discrimination performance on the target minimal pair was assessed before and after training. Results: Mandarin nonmusicians exhibited clear distributional learning (listeners in the bimodal, but not those in the unimodal condition, improved significantly as a function of training), whereas Australian English musicians did not (listeners in both the bimodal and unimodal conditions improved as a function of training). Conclusions: Our findings suggest that veridical perception of lexical tones is not sufficient for distributional learning of nonnative lexical tones to occur. Rather, distributional learning appears to be modulated by domain-specific pitch experience and is constrained possibly by top-down interference.