This is the first variationist sociolinguistic study of Cantonese tone-merger using conversational recordings. These data differ from experimental data exploring tone mergers: the speech is continuous and spontaneous, the tones appear in diverse contexts, and speakers are from both Toronto and Hong Kong. We investigated the status of three reportedly ongoing mergers: T2/T5忍 / 引, T3/T6 印 / 孕, and T4/T6 仁 / 孕. We measured three cues (i.e., mean pitch, pitch at 90% duration of the syllable, and pitch slope) in 12,000+ tokens from thirty-two speakers. Using normalized duration and speaker pitch, mixed-effects models showed that unmerged tones are statistically distinguishable in spontaneous speech, but that two of the three “ongoing-merger” pairs are fully merged, and the third is nearly merged. Analyses included segmental and suprasegmental (i.e., phrasal position, word position, adjacent tones) factors affecting pitch. We found no differences between heritage and homeland speaker samples.