Computerized birth files compiled by the State of North Carolina for the years 1975-1977 were analyzed for omissions and inaccuracies. A wide range in the per cent missing values was found for different data items, from essentially none missing (birthweight, sex, race) to about 20 per cent missing (gestational age, paternal social data). Recorded birthweight showed the expected skewing from a normal distribution. The only demonstrable inaccuracy was in the form of digit preference, probably causing errors of +/- 1 oz (28.3 g). Reported gestational ages were more suspect, falling outside the range of biologically plausible gestation length in 2.8 per cent of cases. An additional 1.5 per cent of gestational ages were found to be misdated by four to 20 weeks based on the observed bimodal weight distributions among births of the same reported gestational age. Hospitals of various sizes and administrative affiliations submitted records with missing or inaccurate gestational age data with roughly equal frequency. These records were found to come from a socio-demographically high-risk subpopulation. The implications of elimination of incomplete or erroneous birth record data in perinatal epidemiologic research are discussed.