Urine and plasma oestriol, plasma progesterone, human placental lactogen, beta 1-glycoprotein and serum cystyl aminopeptidase were measured at intervals during 608 pregnancies. The predictive accuracy of low values for identification of pregnancies with low birthweight outcomes was assessed for each test at various gestations. Data were analysed to obtain 10th-90th centile values for each test from 28 weeks to delivery. Groups with values under different centile levels were compared: those under the lower centiles had higher proportions but smaller absolute numbers of low birthweight infants than those under higher centiles. No test was superior to the others at all centiles and gestations. Biochemical screening of pregnant populations to identify high-risk groups for intensive fetal monitoring has limited potential. If screening is used, the definition of high-risk groups is best achieved by practical rather than statistical criteria. If monitoring facilities are available and well accepted by patients then higher centile 'cut-offs' to define fetal risk may be used than when they are not. Combining any pair of tests with values below the 10th centile did not reduce false positive and negative predictions any more than could be achieved by movement of centiles up or down for a single test.