A multiple regression equation for estimating premorbid Raven's Standard Progressive Matrices (RSPM) scores on the basis of age and the National Adult Reading Test (NART) was derived. A sample of patients with confirmed cerebral lesions was examined to determine the proportion of patients with significant discrepancies between their predicted and obtained RSPM performance. The sensitivity of the procedure was modest (52 per cent), although a comparison with discrepancies between premorbid and current performance on the Wechsler scales indicated that the RSPM was more sensitive to neuropathology. These findings suggest that when NART-intelligence test comparisons are used to diagnose cerebral impairment, the likelihood of a false-negative diagnosis is high.