A difficult problem in English syntactic analysis is the treatment of complex nominals. In the context of text-to-speech one needs not only to syntactically analyze such constructions, but also to assign appropriate stress. I describe NP, a complex nominal analyzer currently under development in the context of the Bell Labs Text-to-Speech system. NP makes use of syntactic, semantic, lexical and statistical information to decide on the most appropriate structure for complex nominals; phonological rules are then applied to produce stress appropriate to that structure. An evaluation of the performance of the current program is presented.