This paper addresses the question: How does implementation influence the effectiveness of mandatory welfare-to-work programs? Data from three large-scale, multi-site random assignment experiments were pooled; quantitative measures of program implementation were constructed; and multilevel statistical modeling was used to examine the relationship between program implementation and effects on short-term client earnings. Individual-level data were analyzed for 69,399 sample members and group-level implementation data were analyzed for 59 local programs. Findings indicate that, other things being equal, earnings effects are increased by: an emphasis on quick client employment, an emphasis on personalized client attention, staff caseloads that do not get too large, and limited use of basic education. Findings also show that mandatory welfare-to-work programs can be effective for many types of people, and that focusing on clients who are especially job-ready (or not) does not have a consistent influence on a program's effectiveness. (Contains 1 figure, 3 tables and 31 footnotes.)