Current research in cognitive modelling has assumed that the interactive nature of processing during language production has been supported by fits to the behaviour of brain-damaged patients. In this paper, several previously proposed theories of word production, all based on the interactive spreading-activation theory of Dell (1986), are evaluated in the context of picture naming. Using a new corpus of data from 50 Italian aphasic patients, we find that patient patterns do not seem to demand an interactive theory. We also explore the corollary ideas of continuity between normal and random performance, and global damage in aphasia, and we find that they are incompatible with our data. We argue that it is very difficult for a modelling study to yield strong constraints on cognitive theory. Although valuable, such efforts currently do not contribute evidence for interactive processing in language production.