ABSTRACT:Objective: To estimate the prevalence of disability and its association with sociodemographic and health characteristics stratified by sex. Methods: This is a cross-sectional study with a probabilistic sample including 4,048 residents aged ≥ 18 years in two health districts of Belo Horizonte (MG), Brazil, during the period from 2008 to 2009. The outcome variable "disability" was established based on self-reported problems in body functions or structures. Sociodemographic characteristics ("sex," "age," "skin color," "marital status," "years of schooling," and "family income") and health ("reported morbidity," "health self-assessment," "quality of life," and "life satisfaction") were the explanatory variables. We applied the multivariate decision tree analysis by using the Chi-square Automatic Interaction Detector algorithm. Results: The overall prevalence of disability corresponded to 10.4% and it was higher in females (11.9%; confidence interval -95%CI 10.2 -13.6) than in males (8.7%; 95%CI 6.8 -10.5). In the multivariate analysis, "age" and "morbidity" in females, and "low educational level" and "poor health self-assessment" in males were the variables that best discriminated disability. Disability self-reporting was more frequent among women of working age (40 to 59 years-old) and with lower incomes, as well as in men with lower educational levels and incomes. With regard to health conditions, the highest disability percentages were seen among subjects of both genders that reported three or more diseases and worsened perception of health. Conclusion: Results reinforce the need for a distinct approach, since women of working age and men with lower educational level are more vulnerable to the occurrence of disability.