BACKGROUND: Ultrasonography, the gold standard for establishing a diagnosis in first-trimester vaginal bleeding, is not always readily avaliable. Medical history and gynaecological examination are then used instead, to make a provisional diagnosis. AIM: To determine the diagnostic value of history taking and physical examination in first-trimester bleeding, to differentiate between patients requiring immediate further diagnostic examination from those in whom an expectant policy will initially suffice. DESIGN OF STUDY: Prospective population-based cohort study. SETTING: Seventy-four general practices in Amsterdam. METHOD: Two hundred and twenty-five patients with first trimester vaginal bleeding were referred for an early pregnancy assessment. The data from 204 patients were analysed. Two diagnostic models were constructed based on symptoms and the results of gynaecological examination to identify diagnostic subgroups relevant to clinical practice. RESULTS: Model 1, which separates viable pregnancies from other diagnoses, increased pre-test probability from 47% to a post-test probability of 70%. Model 2, which enabled the identification of complete miscarriages, resulted in a post-test probability of 41% of a complete miscarriage, given a pre-test sample probability of 25%. The tentative diagnosis of a general practitioner, based entirely on clinical judgement, turned out to be a poor predictor for the ultrasonographically confirmed diagnosis (pre-test probability of 47% changed to a post-test probability of 58%). CONCLUSION: This study shows that, in first trimester bleeding, neither statistical prediction models based on signs and symptoms, nor clinical judgement, are valid replacements for ultrasonographic assessment in establishing a diagnosis.