This study examines the production of third language (L3) German consonant clusters by 26 L1 Cantonese-L2 English bilinguals, with the aim of uncovering the possible cross-linguistic influences on L3 pronunciation. Learners' production of 17 onset and 21 coda German consonant clusters were auditorily analysed with respect to accuracy and error type. Findings showed that L3 pronunciation accuracy was affected by the similarities between each individual L3 consonant cluster and its possible correspondences in L1 Cantonese and L2 English. An L3 German consonant cluster that contained more Cantonese and English consonants would be produced more accurately, whereas an L3 consonant cluster that looked similar to English but mismatched with English orthography would generate production errors (e.g. kn [kn] [right arrow] kn [n]). Individual variation between different L3 consonant clusters suggests that cross-linguistic influences on the L3 can occur from both the L1 and the L2 at the feature level and can be both positive and negative. At the same time, the study also shows that L3 speech is significantly influenced by markedness and language experience factors.