Research evidence relating to the utility of the RMC evaluation models of compensatory education employing non-normed tests is examined. The history and evolution of five early models into the current norm-referenced model utilizing a non-normed test (Model A2), non-normed versions of a comparison model (Model B2), and the regression model (Model C2) are described. Technical considerations of minimum correlation, score overlap and score distribution for Model A2 and the estimation of population standard deviations on the non-normed test for Models B2 and C2 are discussed. Practical considerations include group size, grade level, type of score used and instrument sensitivity. Problems and proposed solutions are discussed. The preference for non-normed tests is shown to stem from a perceived discrepancy in the sensitivity of tests to instructional objectives, so the task of comparing alternative implementation strategies for various models is considered. The utility constraints of the Education Consolidation and Improvement Act of 1981 are discussed. In the light of practical alternatives presented and seemingly insurmountable problems in Models A2, B2 and C2, it is recommended they be abandoned. (Author/CM)