OBJECTIVES: Propensity Score Matching (PSM) is a common method in many retrospective studies to control for differential treatments. PSM controls for variables where patients are selected for one treatment over another based on aspects of their care that are unknown to the researcher or not a part of the study.  This study uses simulated data comparing two cohorts within a population treated for a common psychiatric disorder. Data are analyzed to determine if regression artifacts (RA) are present in the data, uncontrolled by PSM. RA in this context are Type I errors. METHODS: Variables commonly used to diagnose patients with Major Depression were simulated: Age, Gender, Ethnicity, Global Assessment of Functioning, Beck Depression and Beck Anxiety scores. Distributions of N=100,000 were simulated for each variable using population values. From these distributions, samples of n=100, n=250 and n=500 were drawn based on typical values that would be seen in a patient with Major Depression. The outcome measure Dependent Variable was the score on the Beck Depression scale, using success of treatment values from 10-15 percent, and correlated with the pretest score using Chomsky’s decomposition. PSM was used on a ratio of 1:1. Analysis methods were group and paired t-tests as well as a difference in difference analysis at the end of the study. RESULTS: Type I error occurred in each simulation and were correlated with sample size. RA, leading to Type I error were more common at lower sample sizes, in excess of 70%, to a minimum of 54% for n=500. CONCLUSIONS: This study demonstrates that RA occur in basic experiments designed to specify treatment effects. Researchers who use PSM methods need to be aware of situations where RA are likely to occur. Standard statistical controls for RA are being tested to see if they correct for RA and Type I error when PSM is used.