Current proposals suggest that agrammatic speakers have severe deficits in producing and understanding wh-questions. Whereas the CP layer is assumed to be pruned such that wh-questions can no longer be produced, for comprehension it has been suggested that the agrammatic deficit leads to a deletion of movement traces in syntactically derived sentences. We present data from three experiments testing the production and comprehension of wh-questions in nine German agrammatic Broca's aphasics. In an elicitation and a repetition task we find that complete and grammatical wh-questions can be produced by German agrammatic aphasics, thus indicating that the CP layer can still be projected. The results of a picture-pointing task show, however, that deficits in the comprehension of wh-questions (who and which N questions) are common. A closer analysis reveals a variety of error patterns in our subjects. The additive effects that non-canonical word order and the discourse-linking of a referential NP (which N) exert on the comprehension performance of our subjects cannot be captured by current syntactic deficit accounts on agrammatism.