Face recognition can be viewed as part of a divergent set of operations in object recognition, in which information from common low-level visual mechanisms feeds forward into increasingly specialized processes for different object types. This divergence may also involve hemispheric specialization, notably for faces in the right and words in the left hemisphere. However, in person recognition, face processing is one of a set of sensory inputs that converge upon access to stored information about people. We review the literature and evidence from a cohort of acquired prosopagnosic subjects, on three issues concerning selectivity. First, we review the data on object recognition in prosopagnosia, and recent evidence that, after adjusting for pre-morbid car expertise, impairments of car identification are common in our cohort, particularly among car experts. Second, we review the data on word processing in prosopagnosia. In our cohort we show that the word-length effect in single word reading is normal after right-sided lesions, but the discrimination of font and handwriting is impaired in most of our subjects, regardless of lesion location. Third, we discuss the status of voice recognition in prosopagnosia, and show that in our cohort, right anterior temporal lesions do not impair this function, but bilateral ones do. Together, these findings suggest that the processes for faces, cars and visual text involve either the same neural resources or parallel processes in close proximity. Voice and face processing remain distinct in our subjects, and confirm that right anterior temporal lesions cause an associative prosopagnosia rather than a multi-modal person recognition syndrome.