Two severely aphasic patients who made frequent semantic errors in verbal picture naming, as well as frequent errors in written naming, were studied. Contrasting patterns of errors across various language tasks provide evidence that the two patients' naming errors arose from different underlying deficits. The effectiveness of cueing hierarchies on improving each patient's written naming was demonstrated in single-subject experiments using a multiple baseline design. Although both patients exhibited acquisition and maintenance of written naming, only one showed generalization to verbal naming and to untrained stimuli. Different results are interpreted as a reflection of separate sources of the subjects' naming errors. It is concluded that determining the cognitive basis of an individual's naming difficulty may permit predictions concerning language behaviors that are likely to improve concurrently as a function of treatment. Also, reporting specific deficits of patients should allow other clinicians to select appropriate candidates for therapy procedures found to be effective in within-subject treatment experiments.