To estimate the prevalence of blindness and partial sight in the general and diabetic populations and to measure the impact of ethnicity and gender on these rates. Data on visual impairment in the population to be studied was obtained from the Royal Leicestershire Rutland and Wycliffe Society for the Blind (RLRWSB). Using the validated technique of capture-mark-recapture (CMR), a census day was set and data from four sources collected: hospital admissions data, GP returns, wards, and mass media. Computer cross-linkage to Leicestershire Health Authority data ensured patients were alive and resident in Leicestershire at the census date. Estimates of diabetes, ethnicity and gender prevalence were calculated using the 1991 census. CMR calculations revealed that there was under ascertainment of visual impairment in the RLRWSB data. The CMR calculations with direct age standardization estimate the prevalence of blindness and partial sight in the general and diabetic populations to be 51.8/10 000 (CI 50.6-53.0) and 148.8/10 000 (CI 122.1-175.5), respectively. In the diabetic population there was a higher proportion of visually impaired females than males (P < 0.05), with no overall ethnic differences; however, rates were lower than expected in female Indo-Asians. In the general population there was a higher proportion of visually impaired Indo-Asian people (P < 0.01). Our study has shown the prevalence of visual impairment in both the general and diabetic populations and has demonstrated both ethnic and gender differences. CMR allows the calculation of prevalence of blindness and partial sight from existing data. When comparing the prevalence of visual impairment, whether from diabetes or other causes, it is important to know the gender and ethnicity of the 'at risk' population, as well as the rates in the background population.