The fragility of expertise is a known, but little understood, feature of expert reasoning. Essentially, fragili(y refers to the performance degradation of experts as task properties change. A study is presented in which the fragility of expertise in a complex, real-world task -- reactive scheduling -- is investigated. Six novices (students, trained in the task but with no experience in the domain) and three expert schedulers (ranging from six to 20 years of experience in the domain) each completed six reactive scheduling tasks varying in difficulty. All subjects were run individually and their protocols (verbal and action) were recorded on video-tape. Simple modifications to the task environment were sufficient to degrade the pelfonnance of the experts, sometimes to the level of the novices. However, an analysis of the behavior of the subjects suggests that a problem space characterization of fragility can explain how that degradation occurred. The behavior captured in the video-tapes (both verbal utterances and physical actions) show that, in this task, the primary source of degradation was the inappropriate formation of problem space components. That is, experts were "stuck in the wrong problem space." Specifically, the experts would use inadequate search control knowledge while traversing problem spaces and/or repeatedly attempt to implement operators or types of search control knowledge that were not allowed in the experimental task, but were quite valid in the real task setting. We conclude by discussing the concept of expert fragility and how it should be taken into account when designing systems based on the construct of expertise: expert systems.