The accuracy of the urinary group B streptococcal antigen latex agglutination (LA) test for screening infants at risk of group B streptococcal (GBS) sepsis in the first 24 hours of life was prospectively studied in 236 infants for six months. Infection with GBS was defined by a positive blood culture while colonisation was defined by GBS cultured from any other site. The combination of infection and colonisation was used as the gold standard for the LA test. Although the LA test had a sensitivity of 90%, the specificity was only 70%, the positive predictive value 12% and the false positive rate 30%. The overall accuracy was only 71%. The LA test was unable to predict GBS sepsis in infants at risk of the disease. The false positive rate was unacceptably high and could not be potentially accounted for in 11 infants. However, a negative test was useful in excluding GBS disease.