Relative clauses (RCs) are considered islands for extraction, yet acceptable cases of overt extraction from RCs have been attested over the years in a variety of languages: Danish, Swedish, Norwegian, Japanese, Hebrew, English, Italian, Spanish, French, and also in Lebanese Arabic and Mandarin Chinese, where covert extraction from an RC is observed. The possibility for extraction has often been presented as evidence against a syntactic theory of locality, and in favor of constraints defined in terms of information structure, or processing limitations and constraints on working memory. Another possibility, still hardly explored, is that locality is determined syntactically, combined with a more fine-grained structure for RCs and a theory of how extraction from this structure interacts with the theory of locality. I argue in favor of the latter approach. I assume the structural ambiguity of RCs and argue that while externally headed RCs do block extraction, extraction is possible, under certain conditions, from a raising RC, and is formally similar to extraction from an embedded interrogative.