Objective: Our aim was to determine the accuracy of cervical length and funneling of the internal os in the prediction of the spontaneous very preterm birth of twin pregnancies. Study design: In a prospective study at 13 centers, routine transvaginal ultrasound scanning was used to examine 251 women with twin pregnancies at 22 weeks of gestation and 215 women at 27 weeks of gestation; cervical length, spontaneous cervical funneling, and funneling after transfundal pressure were evaluated. Receiver operating characteristic curve analysis was used to determine the best cervical length for the prediction of spontaneous preterm birth before 32 and 35 weeks of gestation. We calculated the sensitivity, specificity, and predictive values for cervical length and for funneling. Results: The median gestational age at delivery was 36 weeks in both populations. Of the population that was included at 22 weeks of gestation, 5.2% (13 women) gave birth spontaneously before 32 weeks of gestation, and 13.2% (33 women) gave birth spontaneously before 35 weeks of gestation; the median cervical length was 40 mm. The receiver operating characteristic curve showed no clear best cutoff point for cervical length. For spontaneous delivery before 32 and 35 weeks of gestation, the sensitivity of cervical length ≤30 mm was 46% and 27%, respectively; the specificity was 89% and 90%, respectively. The sensitivity of funneling was 54% and 33%, and its specificity 89% and 91%, respectively. After multivariate analysis, only funneling remained significant for delivery before both 32 and 35 weeks of gestation. Of the population that was included at 27 weeks, 3.3% (7 women) gave birth spontaneously before 32 weeks of gestation, and 12.4% (26 women) gave birth spontaneously before 35 weeks of gestation; the median cervical length was 35 mm. The receiver operating characteristic curve showed 25 mm to be the best cutoff point for cervical length. For spontaneous delivery before 32 and 35 weeks of gestation, the sensitivity of cervical length ≤25 mm was 100% and 54%, respectively, and the specificity was 84% and 87%, respectively. The sensitivity of funneling was 86% and 54%, and the specificity 78% and 82%, respectively. After multivariate analysis, both indicators remained significant for delivery before 35 weeks of gestation. Funneling after transfundal pressure at 22 or 27 weeks did not predict very preterm delivery. Conclusion: Cervical length and funneling both predict the very preterm birth of twins. Although cervical length is the predictor of choice at 27 weeks of gestation, at 22 weeks of gestation the diagnostic values of both parameters are close. (Am J Obstet Gynecol 2002;187:1596-1604.)