Purpose The optimal approach to diagnosing deep venous thrombosis is not entirely clear. In this prospective cohort study, we aimed to evaluate the yield of two methods of assessing the pretest probability of deep venous thrombosis-the treating physician's implicit assessment and the Wells score, a validated prediction rule that incorporates signs, symptoms, and the presence or absence of an alternative diagnosis-used in isolation and in combination with D-dimer measurement. Subjects and methods We studied 278 patients who were referred for suspicion of deep venous thrombosis. All patients were stratified into groups of low, moderate, or high risk of deep venous thrombosis on the basis of the clinical assessment and Wells score, and underwent rapid quantitative D-dimer testing (with a cutoff of 500 μg/mL), ultrasound examination, and follow-up for the occurrence of venous thromboembolism. Results Eighty-two patients (29%) had a deep venous thrombosis. The accuracy of both methods was good (area under the receiver operating characteristic curve = 0.72), despite only fair agreement at the level of individual patients (weighted kappa = 0.31; 95% confidence interval [CI]: 0.23 to 0.40). The negative predictive value of D-dimer measurement was 96% (95% CI: 91% to 100%). When restricted to patients with low pretest probability, the negative predictive value of D-dimer measurement was 100% (95% CI: 96% to 100%) with the use of the Wells score and 96% (95% CI: 88% to 100%) with the physician's assessment. Our results were unchanged in analyses restricted to patients with proximal deep venous thrombosis or outpatients. Conclusion Clinical assessment to stratify a patient's likelihood of having deep venous thrombosis should be taught to physicians.