Although early ultrasound (<20 weeks' gestation) systematically underestimates the gestational age of smaller fetuses by approximately 1-2 days, this bias is relatively small compared with the large error introduced by last menstrual period (LMP) estimates of gestation, as evidenced by the number of implausible birthweight-for-gestational age. To characterise this misclassification, we compared gestational age estimates based on LMP from California birth certificates with those based on early ultrasound from a California linked Statewide Expanded Alpha-fetoprotein Screening Program (XAFP). The final sample comprised 165 908 women. Birthweight distributions were plotted by gestational age; sensitivity and positive predictive value for preterm rates according to LMP were calculated using ultrasound as the 'gold standard'. For gestational ages 20-27 and 28-31 weeks, the LMP-based birthweight distributions were bimodal, whereas the ultrasound-based distributions were unimodal, but had long right tails. At 32-36 weeks, the LMP distribution was wider, flatter, and shifted to the right, compared with the ultrasound distribution. LMP vs. ultrasound estimates were, respectively, 8.7% vs. 7.9% preterm (<37 weeks), 81.2% vs. 91.0% term (37-41 weeks), and 10.1% vs. 1.1% post-term (>or=42 weeks). The sensitivity of the LMP-based preterm birth estimate was 64.3%, and the positive predictive value was 58.7%. Overall, 17.2% of the records had estimates with an absolute difference of >14 days. The groups most likely to have inconsistent gestational age estimates included African American and Hispanic women, younger and less-educated women, and those who entered prenatal care after the second month of pregnancy. In conclusion, we found substantial misclassification of LMP-based gestational age. The 2003 revised US Standard Certificate of Live Birth includes a new gestational age item, the obstetric estimate. It will be important to assess whether this estimate addresses the problems presented by LMP-based gestational age.