The ability of routine vaginal examinations to improve the prediction of preterm delivery was assessed in a group of 6909 women who were registered at each prenatal visit and on whom this examination had been carried out. We compared two risk scores, one including known risk factors (maternal characteristics and symptoms reported by women), and the other including these factors and the findings of vaginal examination. These risk scores were computed by multiplying the adjusted odds ratio estimations obtained by logistic regressions. The prediction of preterm delivery was improved significantly by vaginal examination at 25 to 28 weeks' and 29 to 31 weeks' gestation. However, the improvement was not very large: when 30% of nulliparous women were classified as high risk at 29 to 31 weeks, the sensitivity was 55% when considering only the risk factors and 63% when adding the findings of vaginal examination; the percentages were 52% and 55%, respectively, for parous women. These results partially explain why the medical practice of routine vaginal examinations varies from country to country.