In this study the authors test whether matching using intact local groups improves causal estimates over those produced using propensity score matching at the student level. Like the recent analysis of Wilde and Hollister (2007), they draw on data from Project STAR to estimate the effect of small class sizes on student achievement. They propose a strategy for intact group matching in which they match treatment cases to control cases in other schools. A secondary goal of this analysis is to determine whether the use of geographic covariates (including latitude and longitude, as well as Census variables) improve the quality of matches over the use of student, teacher, and school covariates alone. They hypothesize that by incorporating physical distance into their matching, they will increase their likelihood of finding maximally similar comparison classrooms.