Objective: To determine treatment noncompliance rates among patients participating in a municipal tuberculosis control program and to identify the variables related to noncompliance depending on the type of treatment strategy used. Methods: A longitudinal non-concurrent cohort study was carried out involving two cohorts of patients participating in the Tuberculosis Control Program of the city of Carapicuíba, Brazil. The first cohort comprised 173 patients with tuberculosis treated from January 1, 2003 to December 31, 2003 using self administration of treatment, and the second comprised 187 patients with tuberculosis treated from July 1, 2004 to June 30, 2005 using the directly observed therapy, short-course strategy. Results: Noncompliance rates decreased from 13.3% (self administration of treatment) to 5.9% (directly observed therapy, short-course), a significant difference (p < 0.05). For the self administration of treatment strategy, the variables significantly associated with treatment noncompliance were as follows: being an unregistered worker (relative risk [RR] = 3.06); retreatment (RR = 2.73); alcoholism (RR = 3.10); and no investigation of contacts (RR = 8.94). For the directly observed therapy, short-course strategy, no variables were significantly associated with noncompliance. Conclusion: The directly observed therapy, short-course strategy decreased noncompliance rates and produced better treatment outcomes, even when the risk factors for noncompliance were the same.