The determinants of change are assumed to be represented by a linear regression model. The implications of using raw difference scores and of using residual change scores are evaluated within the framework of this model. Given the regression model and a determinant of change (Z) other than initial status (X) , use of the above traditional measures of change results in biased estimates of the influence of Z on change when X and Z are correlated. Parallel results are obtained when the measures are treated as error free and when they are treated as fallible.