A major source of bias in health surveys is non-response on the part of those selected to take part in a study. In a survey of the oral health of older adults in Ontario, Canada, we used an initial telephone survey based on random digit dialing with a personal interview and clinical examination follow-up. The telephone survey was completed by 3033 individuals (78.0%) of those sampled. The follow-up was completed by only 907 (30.0%) of these subjects. Nevertheless, there were no major differences in the characteristics of those completing the telephone survey and those subsequently participating in the follow-up. Non-response bias analysis indicated that differences between crude and adjusted estimates of the prevalence of oral conditions were small and the effect of non-response on estimates of the relationship between socioeconomic status and oral health in this population were also small. These results indicate that response rates lower than those conventionally regarded as acceptable do not necessarily compromise the results of epidemiological studies.