In intervention research, it is critical to determine not just if an intervention is effective, but for whom it is effective and "under what circumstances" those effects occur. Moderators can be the key to answering those questions. A moderator is a variable that affects either the direction or the strength of the relationship between the predictor (curriculum condition, in this case) and the dependent variable (here, child outcomes) (Baron & Kenny, 1986). Identifying those variables that help specify the conditions under which interventions are most effective is central to social science research (Cohen et al. 2003). Moderators of curricular effects may be particularly important to scale-up studies. There may be no more challenging educational and theoretical issue than scaling up educational programs across a large number of diverse populations and contexts in the early childhood system in the U.S. The paucity of high-quality instruments, or the use of any measure of the fidelity of implementation, is one of the most important deficits in the field of scaling up educational innovations (Borman et al., 2003). This paper examines possible moderators, using fidelity as the primary variable of interest, of the effects of an early mathematics curriculum used in a scale-up study across three states. Participants in this study included primarily at-risk preschoolers between the ages of 4 and 6 from low-income households. Data from this study demonstrated teacher adoption of many aspects of the curriculum, and the degree of adoption was related modestly to the amount children gained in math knowledge over the preschool year. While teachers enacted more math instructional activities, changing the quality of their instruction was much more difficult. Moreover, the overall instructional classroom environment in treatment and control classrooms remained very similar, as did children's learning behaviors. The results from this scale up investigation demonstrate how difficult immediate and sustained change is to effect. (Contains 2 figures.)