This study examines how students' experimentation with a virtual environment contributes to their understanding of a complex, realistic inquiry problem. We designed a week-long, technology-enhanced inquiry unit on car collisions. The unit uses new technologies to log students' experimentation choices. Physics students (n = 148) in six diverse high schools studied the unit and responded to pretests, posttests, and embedded assessments. We scored students' experimentation using four methods: total number of trials, variability of variable choices, propensity to vary one variable at a time, and coherence between investigation goals and experimentation methods. Students made moderate, significant overall pretest to posttest gains on physics understanding. Coherence was a strong predictor of learning, controlling for pretest scores and the other experimentation measures. We identify three categories of experimenters (intentional, unsystematic, and exhaustive) and illustrate these categories with examples. The findings suggest that students must combine disciplinary knowledge of the investigation with intentional investigation of the inquiry questions in order to understand the nature of the variables. Mechanically executing well-established experimentation procedures (such as varying one variable at a time or comprehensively exploring the experimentation space) is less likely to lead students to valuable insights about complex tasks. Our proposed categories extend and refine previous efforts to categorize experimenters by linking scientific procedures with understanding of the science discipline. (Contains 7 figures and 4 tables.)