We designed a prospective observational study to attempt to validate two recently described clinical decision rules for knee radiography. Consecutive patients aged > or = 15 yr with acute knee injuries occurring less than 1 wk prior to presentation were included for study. Patients with distracting conditions, open knee injuries, or previous surgery were excluded. Each patient was assessed for 7 historical and 15 physical examination criteria that were recorded on a standardized data collection instrument. Radiographs were ordered at the discretion of the attending physician and were read by two board-certified radiologists. When radiographs were not ordered, structured telephone follow-up was performed after 3 wk. The main outcome parameter was the presence or absence of a clinically significant fracture. There were 351 patients in the study; 26 (7%) had knee fractures. Fractures were significantly associated with an increased prevalence for two of the three criteria in the rule derived by Bauer: inability to weight bear immediately or in the emergency department (ED; 76.9% of patients with a fracture vs. 29.8% of patients without a fracture) and effusion (53.8% vs. 28.9%, respectively). Ecchymosis was not significantly associated with fracture (19.2% with fracture vs. 9% with no fracture). Use of the Bauer rule would have led to a radiographic evaluation of 22 of the 26 patients with knee fractures (sensitivity = 84.6%, specificity = 48.9%). Fractures were associated with a significantly increased prevalence for three of the five criteria in the decision rule proposed by Stiell: isolated patella tenderness (30.8% with fracture vs. 14.5% with no fracture), inability to flex the knee to 90 degrees (42.3% vs. 19.7%, respectively), and inability to weight bear immediately and in the ED (57.7% vs. 18.8%, respectively). Age > or = 55 yr (23.1% vs. 12.0%, respectively) and fibula head tenderness (11.5% vs. 5.5%, respectively) were not significantly associated with fracture. Use of the Stiell rule would have led to radiographic evaluation of 22 of the 26 patients with knee fractures (sensitivity = 84.6%, specificity = 49.8%). We conclude that neither clinical decision rule is 100% sensitive. Further refinement will be necessary to identify all patients with knee fractures.