One source of bias, regression artifacts, in program evaluations that use regression discontinuity design is identified, and its importance is discussed. A method is suggested for its correction. Although use of this method does not solve the problem of selection bias, it does improve the reliability of nonexperimental methods. (SLD)