Fodor (1983) has proposed that face perception is carried out by an informationally encapsulated module, whose operation is unaffected by context or expectancies. We tested the modularity hypothesis by examining whether discriminations between normal and distorted versions of famous faces can be primed, either by the name of an associated person (semantic context) or by a valid cue as to the identity of the target face (expectancy). A preliminary experiment showed that, in the absence of priming, discriminations between normal and distorted versions of a face were unaffected by whether the target faces were familiar or not, confirming that these judgments tap perceptual, not postperceptual (semantic), coding processes. In Experiment 1, accuracy was significantly higher when target face pairs were preceded by related name primes, as compared with unrelated ones. In Experiment 2, reaction times were significantly faster for targets preceded by a valid identity cue than for targets preceded by an invalid one. Neither effect could be explained as a speed-accuracy tradeoff. These results fail to support Fodor's conjecture that face processing is encapsulated.