Symptoms and clinical signs individually are inaccurate for the diagnosis of DVT. However, when assessing patients with leg symptoms, clinicians have access to additional information, such as whether or not DVT risk factors are present that could improve the accuracy of clinical judgment. The purpose of this study was to identify which clinical variables best predict DVT, and to use these variables to create a clinical prediction index for DVT. We studied 271 university hospital patients with a first episode of symptomatic, clinically suspected DVT. The prevalence of DVT was 27%, of which 71% were proximal. At baseline, information was collected on demographic features, comorbidity, and symptoms and signs. A Bayesian model selection strategy was used to estimate the logistic regression model that best predicted DVT. Male sex [OR = 2.8 (1.5, 5.1)], orthopedic surgery [OR = 5.4 (2.2, 13.6)], warmth [OR = 2.1 (1.2, 3.9)] and superficial venous dilation on exam [OR = 2.9 (1.4, 5.7)] were independent predictors of DVT. Using the model, a clinical prediction index that categorized patients into different levels of DVT risk was created, and was useful in a theoretical strategy aimed to limit the need for contrast venography in patients with suspected DVT, such that 96% of study patients could have avoided contrast venography. This index should be evaluated prospectively in other patient populations.