Students (n = 1,069) from 60 4th-grade classrooms were taught the control of variables strategy (CVS) for designing experiments. Half of the classrooms were in schools that performed well on a state-mandated test of science achievement, and half were in schools that performed relatively poorly. Three teaching interventions were compared: an intervention that used examples of invalid designs to explain the logic of CVS (Invalid condition); an intervention that used examples of valid designs to explain CVS (Valid condition); and a control condition, in which no explicit instruction was provided until after an immediate posttest. The relative effectiveness of the teaching interventions depended jointly on school achievement level and the type of learning assessment. For students from lower achieving schools, the Valid and Invalid condition were similarly effective and both were superior to the Control condition on an immediate posttest of near transfer. However, only the Invalid condition resulted in better performance than the Control condition on immediate and delayed posttests of intermediate and far transfer. For students from higher achieving schools, the Valid and Invalid conditions were equally effective on the near and intermediate transfer tests. However, on the far transfer test, the Invalid condition was superior to the Valid condition which, in turn, was superior to the Control condition. The pattern of superiority of the Invalid over the Valid condition is attributed to the emphasis in the Invalid condition on the rationale for controlling irrelevant variables.