Direct observation of student study behavior is possible when students use computer-based learning materials. Data recorded on the self-directed studying of more than 100,000 students using a Web-based tool to prepare for U.S. college admissions tests reveal several non-optimal behaviors. Students had a tendency to begin studying within days of the test, to neglect math and science preparation, and to prefer vocabulary drills and short answer verbal questions over questions on reading passages. High school students studying for college admissions showed a different pattern from older students studying for graduate school. A randomized experiment carried out online addressed the possible confound between student preference and Web site design. The data represent the largest sample of admissions test takers ever studied, while also representing the most fine-grained observation of high stakes study behavior ever reported. The relevance of these data for evaluating the causal effect of commercial coaching operations is discussed.