To obtain accurate, unbiased rates of neural tube defects (NTDs) in twins, we conducted a population-based study that included live births and fetal deaths in Los Angeles County, California, ascertaining cases by multiple methods. Twenty-eight twin cases yielded a prevalence-at-birth of 1.6/1,000 twin births, which is significantly higher than the singleton prevalence of 1.1/1,000 births. In twins compared with singletons, the prevalences of both encephalocele and anencephaly are increased, whereas spina bifida is decreased. The twin case male/female sex ratio (.55) is lower than the singleton case sex ratio (.77). Concordance is relatively low at 3.7%, but appears to be higher than recently reported recurrence risks in other low prevalence areas. Stillbirths were most common among female cases and like-sex twins. Our study tends to support proposed etiologic theories associating NTDs with females or monozygotic twins, or both. There is increasing evidence that the etiology of NTDs may differ in high and low prevalence areas. We suggest also that twins and singletons may differ in their response to etiologic factors. The variations among anencephaly, spina bifida, and encephalocele in their association with twinning suggest that there may be different factors that influence the development of each specific NTD. The noted differences among the malformations also indicate that some of the variation among results of other studies of NTDs and twinning may be due to case ascertainment. Including spina bifida cases would decrease the proportion of twins in a study population, while including anencephalics would increase the proportion. Importantly, ascertaining fetal deaths would increase the proportion of anencephalics and case females, so studies of NTDs that do not include fetal deaths will show fewer twins than expected. On the basis of our findings and those of Layde et al., excluding encephaloceles will also decrease the number of twins among NTD cases. When investigating etiologic hypotheses for NTDs, these potential biases must be recognized.