A database of gestation lengths (GLs) was generated over 2 years in three large dairy herds which had used controlled breeding programs (CBPs). This was compared with a database comprising a single calving season of 124 seasonally calving dairy herds using conventional artificial insemination programs (CAIs). Multiphase regression was used to derive two corrected databases by excluding outlying values of gestation length. The mean gestation length derived from the CBP database was slightly shorter than that of the CAI database (280.80 days, n = 775 versus 281.87 days, n = 1986; P < 0.001), but the two databases had similar variances (SD = 4.21 and 4.12 days, for CBP and CAI respectively; P = 0.36). The results from the multilevel analysis showed a mean difference in gestation length of 0.94 (SE 0.50) days; (P = 0.06) between CAI and CBP. The mean gestation length and its SD from the CBP data were used to predict calving curves in herds using CBPs which could be compared with observed calving data. The observed and predicted calving patterns generated for two farms were not significantly different (P = 0.37 and 0.31).The accuracy of the predictions was critically dependent on the completeness and accuracy of conception data and details for cows induced to calve prematurely.