Purpose: Following the demonstration that caffeine effectively competes with oxygen for electrons and also scavenges hydroxyl radicals and singlet oxygen, the differential modification of oxygen-dependent and independent effects of gamma-radiation by caffeine in membranes was examined, using rat liver mitochondria as a model system. Materials and methods: Mitochondria were isolated from the livers of Wistar rats and exposed to gamma-radiation in the dose range of 45-600 Gy (dose rate 15 Gy/min) in the presence or absence of caffeine. To examine the 'oxygen effect', post-irradiation incubation was carried out in the presence of oxygen or nitrogen in buffers saturated wit the respective gases. Membrane damage was examined as lipid peroxidation (assessed as formation of thiobarbituric acid-reactive substances (TBARS), lipid hydroperoxides (LOOH) and conjugated dienes (CD), protein oxidation, depletion of protein thiols, superoxide dismutase or glutathione. Results: Lipid 45 to 600 Gy. Post-irradiation incubation of mitochondria under nitrogen decreased the response, while incubation under oxygen saturation enhanced it significantly. The presence of caffeine during radiation exposure inhibited lipid peroxidation significantly as a function of concentration, in the range of 5 mu m to 4 mM. The inhibition was highest with 4 mM of caffeine. Under oxic conditions, inhibition at 1 mM was significantly more than under anoxia. Anoxia was either ineffective ro marginally increased peroxidation in the presence of caffeine. A similar observation was obtained when membrane damage was assessed as protein oxidation. Radiation-induced depletion of protein thiols was greatly enhanced by oxygen saturation and this was completely prevented by caffeine. This compound also protected against the radiation-induced loss of the antioxidant glutathione and the enzyme superoxide dismutase. Conclusions: The results suggest that caffeine effectively protected membranes against the oxic component of damage but may not do so for the anoxic component.