OBJECTIVES: Investigate whether methods employed to reduce omitted variable bias can be used in a cross-sectional database to identify a relationship between supplemental vitamin usage and the presence of Age-related Macular Degeneration (AMD) in a Latino population. METHODS: Data were obtained from the Los Angeles Latino Eye Study (LALES) and included 6104 subjects. Data were originally collected to assess the prevalence of ocular disease and diabetes. Stepwise logistic regressions and simple logistic regressions for AMD were performed. Propensity scores were then used to control for selection bias. In the first stage, the probability of receiving vitamin supplements for ten years or more was modeled using logistic regression. Subjects were separated into quintiles defined by their propensity scores and we compared the vitamin supplement/no vitamin supplement groups using a two-way analysis of variance model. Finally, the effect of vitamin use on AMD after selection bias adjustment was estimated using logistic regression. RESULTS: In total, 572 cases of AMD were identified. When using stepwise logistic regression, older subjects (OR: 1.043, CI: 1.034-1.052), males (1.624, 1.157-2.280), and people with income level less than $20,000 (1.428, 1.006-2.028) were more likely to develop AMD. Only age and gender were significant using logistic regression and adjusting for potential confounders. The seven covariates found significantly different between the vitamin/no vitamin groups were all non significantly different after adjustment for propensity score quintiles. However, even after selection bias adjustment, vitamin usage continues to be non-significant (0.896, 0.594-1.352). CONCLUSIONS: Although, propensity scores helped reduce potential sources of bias in this cross-sectional database, they did not improve the ability to detect a relationship between vitamin usage and AMD. Other sources of bias, such as the inability to determine the time of development of AMD, unknown dosages and specific vitamins used, could not be addressed by the use of propensity scores.