The relation between PaO2 and retrolental fibroplasia (RLF) was studied prospectively in 719 premature infants born in or treated in the intensive care units of a group of university hospitals. Blood gas studies were performed on 589 of these infants, 66 of whom had a diagnosis of RLF; in 27 of these 66, some grade of mostly nonblinding cicatricial disease developed. The frequency of RLF was highest among infants of lowest birth weight. A multivariate statistical method was used to analyze simultaneously the effect of possible etiologic factors associated with RLF. The occurrence of RLF was found to be unrelated to PaO2, as determined by the limited information available from intermittent sampling. RLF is associated with concentration of oxygen administered in the lightest birth weight group, but the strongest association, aside from birth weight, was with time in oxygen. None of the other variables involving blood chemical values appeared to be associated with RLF. The severity of cicatricial RLF is clearly greater in infants weighing less than 1,200 g at birth. Conservative administration of oxygen may have been responsible for failure to demonstrate quantitative association between PaO2 levels and disease. Agreement between the observed and predicted numbers of infants with RLF demonstrate the strength of the multivariate technique employed in making the statistical analyses.