The Mitchells Plain Disability Survey was undertaken primarily to expand a community-based rehabilitation programme in an underprivileged South African urban community. This descriptive survey used a proportional stratified random cluster sampling strategy (sample size 2424), with stratification by suburb and clusters consisting of 15 adjacent plots. A household screening questionnaire (based on the WHO disability questionnaire), identified people who reported health problems affecting their functional ability, while a second follow-up interview confirmed disablement status and obtained a medical, disablement and demographic profile of the disabled and ascertained their needs. This paper discusses different methodological issues related to the survey design and emphasizes the need for standardization of methods in the disablement field. Sampling issues include sample loss in a multi-staged data collection strategy as well as the non-independence of observations when sampling entire households. The trade-off between studying disability across diagnostic, disablement and age categories, and wide confidence intervals for specific prevalence rates, is discussed. Because of the prohibitive costs validation of disablement status is often omitted in a low-budget project (as this one was), weakening the design of such studies. Even if the 'disabled' are correctly identified, the criteria for identifying respondents determine what type of disablement prevalence will be obtained. Different diagnoses reported on screening yielded different positive predictive values of disability - the most debilitating conditions yielding the highest proportion of disabled people. The quality of the data - evaluated through comparisons of initial and repeat screening interviews, and proxy and self-reporting - is described. There is a need for disability research to continue developing suitable methods for a wide range of purposes. One such is a 'good-enough' survey design which can be implemented rapidly, at relatively low cost, to yield useful results at local level.