A critical analysis is presented of the assumptions that must be made to use the data from aphasia to constrain models of normal language processing. The implication of these assumptions for patient classification and research methodology in aphasia is considered.