The study measured the effect of tested training methods (direct instruction, guided exploration, and a combination of both) on the performance of 497 participants on qualitative and quantitative knowledge tests and control performance task under three complex problem (CP) conditions (easy, medium, difficult). All three methods proved equally suited to meeting the requirements of reproducing qualitative knowledge. For acquisition of quantitative knowledge, the direct-instruction method outperformed a method requiring participants themselves to infer solutions. The greater a CP's intrinsic load due to parallel effects, the more effective direct instruction was at increasing control performance. The guided-exploration group and the combined-method group scored mixed results. The more difficult the CP, the more overall control performance deteriorated regardless of the training method.