Three picture-word interference experiments were conducted to investigate how segmental and tonal information is processed in Cantonese spoken word planning. The picture names were all Cantonese monosyllables with a consonant-vowel-consonant structure. Significant facilitation effects on naming latencies were found when the target (e.g., Symbol. /sing1/, “Star”) and the auditory word distractor (e.g., /ging2/, /sik6/, or /soeng3/) shared two identical phonemic segments (regardless of the segments' syllable-internal position), than when they were unrelated, whereas no reliable effects were obtained when they shared only the vowel (e.g., /dim3/), the coda (e.g., /hung2/), or together with the tone (e.g., /bit1/ or /fung1/). Furthermore, the facilitation effect observed in the consonant + consonant + tone-related condition (e.g., /soeng1/) was reliably larger than that in the consonant + consonant-related condition (e.g., /soeng3/). These results suggest that activation of a single segment is not effective in influencing Cantonese word production and that the lexical tone in Cantonese has a unique role to play. A model which assumes interactivity between lexical and sublexical (including tonal and segmental representations) levels of processing in Cantonese spoken word planning can best account for the available results.