The validity of routine stroke data (official mortality statistics and hospital discharge registries) and a population-based MONICA stroke registry was assessed in a population of 309,806 25- to 74-year-old people in Sweden. The 'true' number of strokes in the population was estimated by screening for non-stroke diagnoses in death certificates and hospital discharge records and by a period of intensified search for otherwise unrecognized non-hospitalized stroke cases. Applying strict stroke criteria, the proportion of false-positive diagnoses was 10% and false-negative 17% in official mortality statistics. Among patients discharged alive from hospital, there were 32% false-positive and 6% false-negative stroke diagnoses. In the MONICA registry, the proportion of false-negative cases was 6% in fatal cases and 4% in non-fatal cases. Diagnostic information for subtyping stroke improved over time in non-fatal cases but remained essentially unchanged in fatal cases. We conclude that official mortality statistics give a reasonably good estimate of fatal stroke cases in Sweden, whereas hospital discharge records reflect poorly the incidence of stroke in the population. The overall quality of the population-based MONICA registry is good, although 4% of all strokes are missed. The proportion of unspecified stroke is substantial and changes over time; this makes longitudinal studies of stroke subtype difficult.