This report examines the effectiveness of four different instructional models used to deliver compensatory education in 1985-86 in Seattle (Washington) by analyzing "sustained gains" in student achievement. A student's "sustained gain" is the reading or mathematics achievement test score gain between spring 1985 and spring 1987, a period that included 1 year of compensatory education in 1985-86. Students may have been served by one of the following instructional models: (1) a traditional "pull out" approach (Pull-Out); (2) computer-assisted instruction (CAI); (3) individual tutoring by community volunteers (HOSTS); or (4) a reading resource specialist (RRS). The sustained gains of compensatory education students were analyzed at the primary and the intermediate levels for each instructional model and were compared with the gains of non-compensatory education students. The following findings are reported: (1) at the primary level, there were no significant differences between gains among any of the models and the comparison group; (2) at the intermediate level, both HOSTS and Pull-Out participants showed reading gains greater than gains for the comparison group, CAI participants showed a negative trend in reading scores, and only Pull-Out participants showed mathematics gains that were greater than the comparison group; (3) the positive effect of the Pull-Out approach duplicates findings of an earlier study; and (4) a study of gains between 1988 and 1990 should permit firm conclusions about the effectiveness of the different instructional models. Statistical data are presented in four tables and two graphs. (FMW)