Previous studies have shown that race and gender are important correlates of survival among patients with cancer of certain sites. Since race and gender influence the stage of disease at diagnosis and the choice of therapy it has been argued that survival differentials may not be real but instead, they represent secondary associations with clinical variables. Therefore, verification of the true prognostic effects of race and gender requires proper controlling for potential confounders, such as stage and treatment. We have studied the 15-year survival experience of a hospital-based cohort of 4527 patients diagnosed with cancer of the mouth over a 28-year period in Brazil. Race and gender were strong predictors of stage and treatment. The odds ratios for no treatment were 1.35 (95% confidence limits [CL]: 1.09, 1.66) for females and 1.63 (CL: 1.29, 2.06) for non-white patients even after adjustment by stage, presumably a key criterion to define treatment. Survival differentials were found for lip cancer, with respect to race, and for cancers of the gum, floor of mouth, and other oral subsites, with respect to gender. Non-whites experienced 2.1 times the risk of lip cancer recurrence (CL: 1.20, 3.61) and 2.3 times the risk of dying from it (CL: 1.29, 4.09) as compared to whites. However, controlling for stage and treatment modality variables by proportional hazards regression reduced the same risk ratios to 1.01 (CL: 0.57, 1.78) and 1.17 (CL: 0.65, 2.13), respectively. The survival advantage experienced by females (17% lower risk of recurrences and 29% lower risk of cancer deaths) regarding other oral sites was independent from the effect of clinical factors.