We studied the effect of penicillin on early-onset Group B streptococcal disease over a 52-month period in neonates who were at high risk of infection. Shortly after birth, 1187 neonates weighing 2000 g or less had blood samples taken for cultures and were randomized into an early-treatment group (given intramuscular penicillin G within 60 minutes of birth) or a control group. The incidence of early-onset disease was 20 per 1000 live births (24 of 1187); the number of infants in the early-treatment group who had disease (10 of 589) was similar to that in the control group (14 of 598). The fatality rates were similar in both groups (6 of 10 vs. 8 of 14). Cultures from blood obtained with one hour of birth were positive in 21 of the 24 infants with disease; 22 of the 24 were symptomatic within four hours of birth. Thus, infection was well established before the first hour of postnatal life. At autopsy, gram-positive cocci were seen in lung sections of four infants in whom cultures of blood obtained after treatment had been sterile; this indicates that giving routine antibiotic therapy before culture samples are obtained can obscure bacteriologic diagnosis. We conclude that penicillin given at birth to neonates weighing 2000 g or less does not prevent early-onset streptococcal disease or reduce excess mortality associated with disease.