Oxygen is determined by irradiating powdered rock samples with 14-MeV neutrons in a dual-transfer system and counting the 6.1- and 7.1-MeV gammas of 16N and their escape peaks. Corrections for beam instability, tritium target decay and spottiness, and electronic drift are made by consecutively switching samples to be irradiated and counted at opposite stations. Corrections must be made for the fluorine content of rocks. Reagent chemicals can be used as standards. The method is very precise, and interferences in the high-energy gamma region are undetectable. Two sets of data are given: uncorrected random results, with statistical evaluation of the precision of the method, and recommended results obtained from reanalyzing samples that showed standard deviations of 0.25% oxygen. Analysis of variance for rock samples shows that the USGS standards can be considered homogeneous for oxygen by this technique.