Three approaches to the neuropsychology of cognitive function are distinguished: the neuroanatomical (where the primary concern is to correlate particular disorders of cognitive function with particular lesion sites), the `general-cognitive' (in which associations are sought between impairments of performance on specific cognitive tasks and general disorders of broadly defined cognitive processes) and the model-building (in which one attempts to interpret the pattern of impairments and preservations of some cognitive function produced by brain damage in terms of an explicit model of the normal operation of this function). I claim that the model-building approach to the neuropsychology of cognitive function must take precedence over the other two. One reason for this is that any disorder of cognitive function can only be defined with reference to some model of that function. I illustrate this claim with reference to acquired disorders of reading, describing current work of a psycholinguistic nature dealing with two acquired disorders of reading: phonological dyslexia and surface dyslexia. A psycholinguistic account of normal reading is used as a theoretical framework to define and to explain the patterns of deficit and preservation observed in these two dyslexias. The detailed account of surface dyslexia in English provided by this framework is then used to make predictions about the nature of surface dyslexia in other languages: alphabetically written languages where all words are regularly spelled, or where homophones cannot occur, as well as ideographically and syllabically written languages. A case of surface dyslexia in an English-Spanish bilingual, in which such predictions were confirmed, is described.