To compare the dates of delivery predicted by last menstrual period (LMP), crown-rump length (CRL) and biparietal diameter (BPD) with the actual date of delivery in a population of pregnant women divided into those with certain and those with uncertain LMP. Healthy women were enrolled at the first visit during their pregnancy to a general practitioner in Odense, Denmark, and underwent ultrasound examinations in the first and second trimesters. Data from a study of 798 women who gave birth in the period August 2001 to April 2003 are presented, although only the 657 spontaneous deliveries were used for analysis (n = 339 and 318 in the certain and uncertain LMP groups, respectively). Data on pregnancy and delivery were collected from the medical records. Wilcoxon's signed rank test was used to test the hypothesis of no difference in prediction error (predicted - actual date of delivery) between the three methods. The median prediction errors estimated by sonography in the first and second trimesters and by corrected LMP according to cycle length were 2.32, 0.16, and 3.00 days, respectively, in women with certain LMP, and 1.71, 0.00, and 3.00 days, respectively, in women with uncertain LMP. The median gestational age at delivery estimated by sonography in the first and second trimesters and by corrected LMP according to cycle length was 282, 280, and 283 days, respectively, in both groups. An ultrasound examination in the second trimester (17-22 completed weeks) is the best predictor of the date of delivery at the individual level, followed by an ultrasound examination in the first trimester. Having an uncertain LMP does not affect the sonographic prediction of date of delivery.