To establish the relationship of cervical length at 23 weeks of gestation to the risk of spontaneous delivery before 33 weeks and to determine the possible additional risk if funneling is present. During a 36-month period, 6819 women with singleton pregnancies underwent transvaginal sonographic cervical assessment at 22-24 weeks as a screening test for preterm delivery. The distribution of cervical length and the prevalence of funneling, defined as dilatation of the internal os of > or = 5 mm in width, were established. Women who underwent cervical cerclage, iatrogenic preterm delivery or were lost to follow-up were excluded from further analysis. In the remaining 6334 pregnancies, logistic regression was used to examine the contribution of cervical length and funneling to the risk of spontaneous preterm delivery before 33 weeks. The median cervical length was 36 mm and in 1.6% of cases the length was < or = 15 mm. There was a significant inverse association between cervical length and percentage rate of spontaneous delivery before 33 weeks. Funneling of the internal os was present in about 4% of pregnancies and the prevalence decreased with increasing cervical length from 98% when the length was < or = 15 mm to about 25% for lengths of 16-30 mm and less than 1% at lengths of > 30 mm. The rate of preterm delivery was 6.9% in those with funneling compared to 0.7% in those without funneling (chi2 = 86.7; P < 0.0001). However, logistic regression analysis demonstrated that funneling did not provide a significant additional contribution to cervical length in the prediction of spontaneous delivery before 33 weeks (odds ratio for short cervix = 24.9, Z = 4.43, P < 0.0001; odds ratio for funneling = 1.8, Z = 0.84, P = 0.40). In the prediction of preterm delivery, funneling does not provide any significant contribution in addition to cervical length.