The problem of change scores' correlation with initial status and the problem of low reliability in the measurement of change are addressed. By treating the correlation between initial status and change as a design problem rather than a statistical issue, research questions can be formulated in terms of changes in the shapes of growth curves rather than in terms of inter-group differences in change rates. With the recent developments in Time Service Analysis and in the analysis of cross-longitudinal growth curves, the dependent variable remains simple, thus making the interpretation of any shifts in pattern straightforward. The more stable the test measure is over time, the less chance there is of finding a significant effect. From this perspective, only tests which are unstable yet internally consistant will yield sensitive measures of change. If the low reliability is defined as a measurement problem, then it is not something inherent in the change score itself. By re-defining this and other problems associated with measuring change, progress can be made in solving them. (PN)