66 infants with retinopathy of prematurity (ROP) were compared with 66 infants without ROP. The pairs were matched for gestational age (GA), birthweight, birthdate, and neonatal unit. According to matched pair statistics the following variables were significantly increased with ROP: multiple birth, duration of oxygen administration, total hours of FIO2>0.4, events of paO2>100 torr, paCO2>50 torr, paCO2> fluctuations, pH<7.25, blood transfusions, and duration of artificial ventilation. Using the same set of variables in a multiple regression analysis, several of the variables did not contribute to the variance of ROP. Variables which were eliminated in a backward regression analysis were duration of oxygen administra-tion, number of events of paO2>100 torr, paCO2>50 torr, paCO 2 fluctuations, blood transfusions and birth weight. Unidimensional statistical procedures are inappropriate for the definition of risk factors in prematurity. When the multiple regression equations were calculated separately for infants below (N=73) and above 32 weeks of GA (N=49) the partial regression coefficients of certain variables differed considerably between the two groups. This was especially true for number of events of paO2>100 torr. Using an adequate set of variables and calculating MLR separately for groups of different GA, an early estimation of the retinal risk seems feasible.