The purpose of this paper is to develop a model of meta learning in acquiring expertise in the field of insight problem solving. We gave 4 participants a series of 12 geometric insight problems and analyzed the process microscopically. By modeling causal relationships between observed variables using covariance structure analysis, we found that 1) an increase in both variety of trials and appropriateness of evaluation improves performance, 2) the variety of trials and appropriateness of evaluation are enhanced by activating cognitive coordination with external resources; that is, epistemic interaction with external resources to distribute the cognitive load to the environment and find information that is hidden or hard to compute mentally. These findings suggest that meta learning is a result of the activation of the interaction between the cognitive component (variety of trials and appropriateness of evaluation) and the situational-perceptual component (cognitive coordination with external resources).