This study explored the assumption that the percentile rank of a group of students should be expected to remain the same on a post-test as on a pretest when the treatment of the group in the intervening period was the same as the treatment of the norms population. This assumption is basic to the use of the norm-referenced model for the assessment of Elementary Secondary Education Act Title I programs. Among the effects which may weaken the usefulness of the assumption are sampling errors, floor effects, out-of-level testing effects, selective drop out, poor match between the curriculum and test content, and regression effects. In this study, longitudinal data were studied for up to four years for Portland, Oregon, public school students selected to be below certain specified pretest percentiles. The differences from one year to another range from zero to nine percentiles. This variation may well be large enough to overshadow Title I effects. (Author/CTM)