This study examined the role of the Sonority Sequencing Principle (SSP) in the production of word-initial clusters by Chinese speakers. We conducted an imitation experiment in which Chinese participants had to imitate “model” speech stimuli of the form C1C2a or C1əC2a, with 3 types of sonority profile for C1C2: rising (e.g., kla), plateau (e.g., kpa), falling (e.g., lka). If the SSP influences the production of these clusters, one would expect a higher incidence of vowel insertion for more marked sonority profiles. Our results are consistent with this prediction: more epenthetic vowels were produced within more marked C1C2 clusters, suggesting SSP effects in their production. The acoustic characteristics of the epenthetic vowels suggest they were all the more “intended” (i.e., targeted) that the model clusters were marked. This pattern suggests that the observed SSP effects in terms of incidence of vowel insertion do not solely reflect perceptual effects during the imitation task.