A cross-modal priming technique was used to look for activation of discourse antecedents by three kinds of subject pronominals available in Spanish. Despite strong pragmatic clues there was no indication of antecedent assignment to PRO, subject of a nonfinite verb. Little pro, by contrast, while also a phonologically null element, induced a significant plausibility effect, indicating activation of the pragmatically plausible antecedent, and exhibiting a pattern similar to that of overt pronouns. These data support a modular model of language processing in which the influence of pragmatic inference does not precede that of syntactically-driven information, where available.