The relationship between statistically and clinically significant change has been enigmatic. Jacobson and Truax (1991) have proposed an important step toward rapprochement. However, their suggested index of clinically significant change neglects possible confounding of improvement rate estimates by regression to the mean. An alternative method is described that incorporates an adjustment that minimizes this confounding when statistical regression has been shown to be present. If regression is not present, the Jacobson and Truax method is more appropriate; if regression is present, the Edwards-Nunnally method (Edwards, Yarvis, Mueller, Zingale, & Wagman, 1978) is more appropriate. The two methods are compared, and the effects of instrument reliability and sample deviance on estimated improvement rates are demonstrated using general well-being test-retest data from a sample of older adult mental health outpatients.