OBJECTIVES: The use of registry databases and indirect comparisons has become important in health economic evaluations. Lack of randomization could lead to selection bias due to pretreatment differences between patients. To control for selection bias, the propensity score method (PS) (Rosenbaum & Rubin, 1983) is often applied. However, average treatment effects can vary within different subgroups. It is yet unclear how to perform subgroups analyses when the propensity score method is applied. METHODS: A Monte Carlo simulation is conducted to test the performance of eight different forms of the PS in subgroup analyses. The PSs differ in whether the variables included in the PS were indicators of the subgroup and were related to treatment assignment, to outcome or related to both assignment and outcome. Furthermore the PS is estimated in two ways, primary on treatment assignment only and secondly on a combination of the treatment assignment and subgroup variable. These PSs were used as adjustment in a regression model. Simulations are accomplished for 18 different settings varying sample size, correlation between independent variables and correlation between independent variables and subgroups. RESULTS: The PS without inclusion of the variable for subgroups, but with inclusion of variables related to outcome, is the most appropriate. The PS should be included as a covariate in a regression model together with the variable for subgroups as covariate, where the PS is based on treatment assignment only. Larger sample sizes gave less biased results, while a higher correlation between the independent variables resulted in more biased estimates of the treatment and subgroup effect. Correlation between the independent variables and the subgroup variable did not lead to biased results. CONCLUSIONS: The results show the feasibility and validity of the PS in subgroups analyses when analyzing registry databases and indirect comparisons in economic evaluations.