Objective: To compare last menstrual period and ultrasonography in predicting delivery date. Methods: We used ultrasound to scan 17,221 nonselected singleton pregnancies at 8–16 completed weeks. The last menstrual period (LMP) was considered certain in 13,541 and uncertain in 3680 cases. The duration of pregnancy from the scan to the day of spontaneous delivery was predicted by crown-rump length, biparietal diameter (BPD), and femur length (FL) using linear regression models, and the results were compared with estimates based on LMP. Results: At all gestational ages, ultrasound was superior to certain LMP in predicting the day of delivery by at least 1.7 days. When deliveries before 37 weeks were excluded, crown-rump length measurement of 15–60 mm (corresponding to 8–12.5 weeks) had the lowest prediction error of 7.3 days. After that time, BPD (at least 21 mm) showed a similar error (7.3 days) and was more precise than crown-rump length. Femur length was slightly less accurate than crown-rump length or BPD. Regression models using a combination of any two or three ultrasonic variables did not improve accuracy of prediction. When ultrasound was used instead of certain LMP, the number of postterm pregnancies decreased from 10.3% to 2.7% (P < .001). Conclusion: Ultrasound was more accurate than LMP in dating, and when it was used the number of postterm pregnancies decreased. Crown-rump length of 15–60 mm was superior to BPD, but then BPD (at least 21 mm) was more precise. Combining more than one ultrasonic measurements did not improve dating accuracy.