The effect of the choice of maternal age-specific prevalence curve on the model predicted Down syndrome detection rate was examined. All 19 published regression curves from 11 birth prevalence series in four meta-analyses were included. The detection rate for a five per cent false-positive rate was estimated for three combinations of markers. For free human chorionic gonadotropin and -fetoprotein the lowest predicted detection rate was 62โข3 per cent and the highest 64โข1 per cent, a range of 1โข8 per cent. When unconjugated oestriol was added as a third marker it was 65โข6-67โข3 per cent, a 1โข7 per cent range, and when inhibin A was the fourth marker the detection rate was 72โข0-73โข4 per cent, a 1โข4 per cent range. The number of series included in the regression had the biggest effect: when the authors had used a subset thought to have the highest ascertainment the predicted detection rate generally increased. The type of regression equation used and restrictions on the age range over which the regression was performed were less important factors. The effect of the choice of curve on the predicted increase in detection achieved by incorporating additional markers was relatively small: 3โข1-3โข3 per cent for unconjugated oestriol and a further 6โข1-6โข5 per cent for inhibin A. This analysis shows that the model inaccuracy caused by the maternal age curve is not small but is unlikely to be large enough to influence Down syndrome screening policy decisions. 1998 John Wiley & Sons, Ltd.