This study explores how novices' conceptions about electric circuits affected their self-directed experimentation in a computer-based circuit laboratory. The participants, 22 undergraduates with no formal college instruction in physics, worked individually on an initial problem to assess their conceptual models of circuits and in subsequent sessions on open-ended experimentation with the computer laboratory, attempted to rediscover the laws of electric circuits. Analysis of the initial problem identified four conceptual models in students' solutions. Lower level models were local, piecemeal, and represented the surface features of the materials, whereas higher level models were progressively more cohesive, integrated, and principled. Relations were found between students' causal models and their learning gains in the computer laboratory. Good and poor learners with the laboratory were identified; these groups showed strategic differences in goal-directed planning, generating and interpreting evidence, and managing data. In general, sophisticated models in the initial problem were related to sophisticated reasoning in the computer discovery task. It is suggested that research oriented to the improvement of science learning should attend to interactive relationships between prior conceptions, inference strategies, and the structure of domain-specific knowledge.