The effect of repeat testing in maternal serum multiple marker screening for Down's syndrome was estimated using samples stored in an antenatal serum bank. Human chorionic gonadotropin (hCG) and unconjugated oestriol (uE3) levels were determined in 142 pairs of routinely collected samples which had already been tested for alpha-fetoprotein (AFP). For each marker, about two-thirds of the pairs of values were within 20 per cent of each other and most were within 40 per cent. A multivariate Gaussian model was used to estimate the detection and false-positive rates for different repeat testing policies. A policy of repeat testing those with a high risk of a Down's syndrome term pregnancy given age and marker levels would reduce the false-positive rate but there would also be a reduction in the detection rate. For example, using all three markers and a 1 in 250 cut-off risk, the estimated false-positive rate would fall from 5.3 to 3.8 per cent but the detection rate would decrease from 58 to 55 per cent. A policy of repeating those with either high or borderline risks would produce a modest improvement in screening efficiency. Repeating the 11 per cent with a risk exceeding 1 in 500 yields an estimated false-positive rate of 5.0 per cent and a detection rate of 60 per cent. A policy of selective repeat testing is not recommended as it would not substantially improve screening efficiency. Nonetheless, if a repeat test has been performed, the parameters given in this paper will enable an unbiased estimate of the Down's syndrome risk to be calculated for individual women.