A diversity of user goals and strategies make creation-oriented applications such as word processors or photo-editors difficult to comprehensively test. Evaluating such applications requires testing a large pool of participants to capture the diversity of experience, but traditional usability testing can be prohibitively expensive. To address this problem, this article contributes a new usability evaluation method called backtracking analysis, designed to automate the process of detecting and characterizing usability problems in creation-oriented applications. The key insight is that interaction breakdowns in creation-oriented applications often manifest themselves in backtracking operations that can be automatically logged (e.g., undo and erase operations). Backtracking analysis synchronizes these events to contextual data such as screen capture video, helping the evaluator to characterize specific usability problems. The results from three experiments demonstrate that backtracking events can be effective indicators of usability problems in creation-oriented applications, and can yield a cost-effective alternative to traditional laboratory usability testing.