Contradictory evidence has been reported on the effects of discovery learning approach and the role of instructional explanations. By manipulating the presence of instruction (verbal explanation) and transparency of problem structures, we investigated how effects of instructional explanations differed depending on the transparency of problem structure. We developed an auxiliary representational task that made certain aspects of the problem structure more transparent. Instruction proved irrelevant to those aspects of the problem made transparent by the representation but facilitated learning of those aspects that were left obscure. We suggest that the critical role of instruction is not specifying the steps in solving a problem, but rather making salient the features that are critical to students ability to infer the steps from examples.