To act efficiently in the classroom, teachers need to be able to judge the difficulty of problems from a novice’s perspective. However, research suggests that experts use their own knowledge as an anchor, adjust estimations for others to their own knowledge and thus underestimate the difficulty that a problem may impose on novices. Similarly, experts should underestimate the benefit for novices of task designs derived from Cognitive Load Theory (CLT), as – following the expertise reversal effect – these should be rather disadvantageous for experts. We investigated pre-service and in-service teachers’ competencies in estimating the difficulty of mathematical tasks for novices. Thirty-four pre-service teachers and thirteen experienced teachers solved tasks that varied in instructional design (optimized for novices following CLT versus non-optimized). Participants solved each task and then estimated how many students of a fictional 9th grade class would be able to solve that task. Solution frequencies were collected from fifty-two 9th grade students. In both expert groups, overestimation was clearly more pronounced for non-optimized than optimized tasks, suggesting an expert blind spot that can be explained in terms of an expertise-reversal effect. The experts failed to adequately take into account the benefits of didactic task variation for novice learners. However, whereas pre-service teachers’ overestimations of student performance were large and significant both for non-optimized and optimized tasks, in-service teachers’ overestimations were generally small and failed to approach statistical significance. In contrast to pre-service teachers, in-service teachers seem to have a better mental model of what a student is able to achieve, thus making better judgments of student performance.