Background: The strategy of the Family Health Program has been used as an alternative scenario for prevalence studies. This study intended to present the protocol of the Digitalis Study (DS), prevalence study of chronic diseases, to assess sources of possible selection bias and estimate their impact on the prevalence of self-reported hypertension, diabetes, and myocardial infarction. Methods: Randomization was performed between 38 160 registered individuals with 45 to 99 years by the Family Health Program .Differences between the sources of selection bias (non-acceptance, non-attendance, substitutions) were observed for gender and age. Results: Of the 1,190 residents contacted, 67.1% agreed to participate. There were 144 residents who were not randomly selected but whose participation was confirmed (substitutes). Women and individuals in the intermediate age groups and the prevalence of hypertension were higher among substitutes compared with the randomly selected individuals. Conclusion: The approach of the DS was adequate for the purposes of estimating prevalences, but there was a significant percentage of non-participation. The randomization strategy did not assume outdated records; alternative schedules for visits were not provided for; follow-up at the invitation stage was not sufficient to prevent substitutions and the inclusion of substitutes with a higher prevalence of hypertension.