This paper examines the production of /K/-clusters in a corpus of spoken French. We compare word-final CK#, word-final KC#, and word-initial #CK in the LOCAS-F corpus, which is one of the few French corpora with prosodic annotation [1]. Three production outcomes are compared (K-deletion, schwa insertion, and canonical pronunciation of the clusters), as a function of the following factors: prosodic boundaries, post-lexical context, speech style, and word frequency. Results reveal that for word-final CK#, 17% of the /K/s are deleted, and 39% of the time a schwa is inserted. Post-lexical context, prosodic boundaries, speech style and word frequency all have significant impact on the realization of CK# clusters. The presence of a major prosodic boundary significantly disfavours K-deletion in wordfinal CK#. For word-final CK# and KC#, post-lexical consonantal contexts favour non-canonical realizations. Interestingly, for word-final KC#, epenthetic schwa is still observed in 20% of the cases, and post-lexical context has a significant influence on the insertions. In word-initial #CK, the canonical pronunciation is systematically observed.