Cantonese has six lexical tones (T), but some tone pairs appear to be merging: T2 [25] vs. T5 [23], T3 [33] vs. T6 [22], and T4 [21] vs. T6 [22]. Twenty-eight merging participants and thirty control participants in Hong Kong were recruited for a perception experiment. Both accuracy rate and reaction time data were collected. Seventeen merging participants also participated in a production experiment. Predictive discriminant analysis of the fundamental frequency data and judgments by native transcribers were used to assess production accuracy. Results show that the merging participants still had six tone categories in production, although their “tone space” was more reduced. Tones with lower type frequency were more prone to change. The merging group was significantly slower in tone perception than the control group was. In illustrating the patterns of the ongoing tone merging process in Cantonese, this study contributes to a better understanding of the forces of sound change in general.