In this article we suggest a method of evaluating learner expertise based on assessment of the content of working memory and the extent to which cognitive load has been reduced by knowledge retrieved from long-term memory. The method was tested in an experiment with an elementary algebra tutor using a yoked control design. In the learner-adapted experimental group, instruction was dynamically tailored to changing levels of expertise using rapid tests of knowledge combined with measures of cognitive load. In the non-adapted control group, each learner was exposed to exactly the same instructional procedures as those experienced by the learner's yoked participant. The experimental group demonstrated higher knowledge and cognitive efficiency gains than the control group.