Two previous meta-analyses of the test-coaching literature have yielded conclusions that differ with respect to aptitude test scores. This study attempts to resolve differences in the conclusions of these two analyses by J. A. Kulik, R. L. Bangert-Drowns, and C. L. C. Kulik (1984) and K. Pearlman (1984) by using the methods of F. L. Schmidt and J. E. Hunter, which involve removal of variation due to sampling error, but modifying the basic procedures slightly to use formulas appropriate to the design of each study. Results suggest that small score gains might be expected from coaching and that high-stakes admissions tests are somewhat less amenable to coaching effects than are general intelligence tests. Within each of these groups, greater score gains were found when a pretest was administered in connection with the coaching program. The importance of both adjusting for sampling error (as did Hunter and Schmidt) and using design-appropriate variance formulas (as did Glass) is affirmed. Seven tables present study data. Contains 43 references. (SLD)