A new method is described for calculating maternal serum marker distribution parameters which will improve risk estimation when screening for Down's syndrome. The approach is to calculate parameters using data from the local screened population and data obtained by meta-analysis from all published studies. The local data are used to derive the variance and covariance in unaffected pregnancies. The meta-analysis is used for the mean level in Down's syndrome pregnancies together with the differences in variance and covariance between affected and unaffected pregnancies. Forty-four published studies were analysed. The mean level for Down's syndrome in multiples of the normal median was 0·73 for alpha-fetoprotein (AFP) in total of 1140 pregnancies, 0·73 for unconjugated oestriol (uE3) in 613, 2·02 for human chorionic gonadotropin (hCG) in 850, and 2·30 for free β-hCG in 477. For all four markers, the variance in Down's syndrome was higher than in unaffected pregnancies; for AFP and uE3, the covariances were also higher in Down's syndrome, but for the other markers they were lower. The method was illustrated using data from 6387 pregnancies screened in Leeds.