Recursion is a difficult concept in introductory computer science courses. This study advances our understanding of students’ learning difficulties by applying the Knowledge in Pieces theory to explain students’ inconsistent performance on recursive programming. The theory posits that different task features differentially constrain the behaviors of intuitive knowledge, thus render it productive or problematic in different cases. This subtle mechanism gives rise to the inconsistent performance across tasks that target the same concept. Data were collected using carefully designed test items that represent various levels of situational constraint and also triangulated through follow-up in-depth interviews. The results support the hypothesis. This new understanding creates the opportunity to reconceptualize assessment and instruction practices.