The patients registered with a general practice are usually spread over many census areas and overlap with the distribution of neighbouring practices, so a validated method of aggregating census data to describe the characteristics of practice patients is required. Four methods were used to provide estimates of the percentage of patients aged 75 years and over from census data for 81 practices in Suffolk, England, and these were compared with values derived from the FHSA patient register. Census values for practice areas produced better estimates than those based on the location of the surgery, but the best methods were based on patient-weighted averages of ward and enumeration district data. The finer geographical detail of enumeration districts did not produce substantially more accurate estimates than the ward-level data: both gave estimates with limits of agreement within 2% of the patient register values. Errors in the census, errors in patient registers and selective geographical distributions of practice patients prevent close matching of census and register measures, but two of the methods tested produced estimates that allow broad comparisons between practices.