Recent theories of semantic memory suggest a subdivision into several separate domains of knowledge. The present study examined the structure of semantic person knowledge by analyzing both behavioral and ERP correlates of associative priming (via co-occurrence and/or shared semantic information) versus purely categorical priming (via shared occupational information). Participants performed familiarity decisions for target faces, which were preceded by sandwich-masked prime names at either short (33 msec) or long (1033 msec) SOAs. Although masking effectively prevented explicit prime recognition, faster RTs were generally observed for both associative and categorical priming (relative to an unrelated prime-target condition). At the short SOA, both associatively and categorically primed targets similarly elicited more positive going ERPs compared with unrelated targets in the N400 time range (N400 priming effect), suggesting a common initial mechanism mediating both forms of priming. By contrast, at the long SOA, a typical N400 priming effect was observed for associative priming only, whereas the corresponding effect for categorical priming was small and restricted to a left parietal site. This hitherto unreported interaction of relatedness, and SOA in the N400 suggests an initial fast spreading of activation to a wide range of related targets, which subsequently focuses to more closely related people at longer SOAs. This ability of ERPs to trace the neural dynamics of activation for different forms of prime/target relatedness can be exploited for testing different models of semantic priming.