Fifty spontaneously breathing, preterm infants 48 h old, of 32 weeks' gestation or less, were assigned randomly to receive caffeine citrate (loading dose 20 mg/kg, maintenance dose 10 mg/kg per day) or a placebo (NaCl 0.9%). The study hypothesis was that caffeine reduces the proportion of infants with recurrent hypoxaemic episodes (decrease in transcutaneous PO2 of 20% within 20 s) from 50% to 25%. Transcutaneous oxygen tension (tcPO2) and heart rate were recorded continuously for 50 h and analysed by computer. The two groups were similar in gestational age, birth weight, delivery mode, sex distribution, and Apgar scores. The mean serum concentration (+/- SD) of caffeine 2 h after the second maintenance dose was 96.0 (+/- 34.5) mumol/l in the group receiving caffeine and 9.3 (+/- 12.8) mumol/l in the group receiving a placebo. The mean proportion of infants with more than six hypoxaemic episodes per 12 h in the caffeine groups was higher (57%) than in the control group (51%). The mean proportion of infants with more than six episodes of bradycardia per 12 h was not statistically different in the caffeine group (79%) from the control group (86%). Our results suggest that prophylactic caffeine has little if any effect on the risk of developing hypoxaemic episodes and bradycardia in small preterm infants and the supposed 50% reduction which was considered clinically important at the start of the trial can be rejected with confidence.