Little is known about optimal application and behavior of exposure propensity scores (EPS) in small studies. In a cohort of 103,133 elderly Medicaid beneficiaries in New Jersey, the effect of nonsteroidal antiinflammatory drug use on 1-year all-cause mortality was assessed (1995-1997) based on the assumption that there is no protective effect and that the preponderance of any observed effect would be confounded. To study the comparative behavior of EPS, disease risk scores, and "conventional" disease models, the authors randomly resampled 1,000 subcohorts of 10,000, 1,000, and 500 persons. The number of variables was limited in disease models, but not EPS and disease risk scores. Estimated EPS were used to adjust for confounding by matching, inverse probability of treatment weighting, stratification, and modeling. The crude rate ratio of death was 0.68 for users of nonsteroidal antiinflammatory drugs. "Conventional" adjustment resulted in a rate ratio of 0.80 (95% confidence interval: 0.77, 0.84). The rate ratio closest to 1 (0.85) was achieved by inverse probability of treatment weighting (95% confidence interval: 0.82, 0.88). With decreasing study size, estimates remained further from the null value, which was most pronounced for inverse probability of treatment weighting (n = 500: rate ratio = 0.72, 95% confidence interval: 0.26, 1.68). In this setting, analytic strategies using EPS or disease risk scores were not generally superior to "conventional" models. Various ways to use EPS and disease risk scores behaved differently with smaller study size.