Despite recognition that estimation of gestational age (GA) based on maternal recollection of the last normal menstrual period (LNMP) is fraught with error, it is not generally appreciated that the magnitude and direction of this error vary as a function of the LNMP estimate. Early second-trimester (16 to 18 weeks) ultrasound determinations of the fetal biparietal diameter were used as the "gold standard" to test the validity of LNMP-based GA estimates in 11 045 women. The large majority of deliveries occurring at or near term showed LNMP estimates that were valid within plus or minus seven days of the ultrasound estimate. As the LNMP GA deviated progressively toward earlier or later GAs, however, the discrepancies became quite marked, especially for postterm dates. The positive predictive values of the LNMP GA estimates decreased dramatically from term (.949) to preterm (.775) to postterm (.119) deliveries. These systematic errors in menstrual GA estimates have profound implications for unnecessary induction, dysfunctional labor and cesarean section, and resultant neonatal and maternal morbidity.