The goal of this study was to evaluate the potential of using the difference between the 1H NMR frequencies of water and N-acetylaspartic acid (NAA) to measure brain temperature noninvasively. All water-suppressed and non-water-suppressed 1H NMR spectra were obtained at a field strength of 4.7 T using a surface coil. Experiments performed on model solutions revealed a decrease in the difference between NMR frequencies for NAA and water as a linear function of increasing temperature from 14 to 45 degrees C. Changing pH in the range 5.5-7.6 produced no discernible trends for concurrent changes in the slope and intercept of the linear relationship. There were minor changes in slope and intercept for solutions containing 80 or 100 mg of protein/ml versus no protein, but these changes were not considered to be of sufficient magnitude to deter the use of this approach to measure brain temperature. The protein content of swine cerebral cortex was found to remain constant from newborn to 1 month old (78 +/- 12 mg/g; n = 41). Therefore, data collected for the model solution containing 80 mg of protein/ml were used as a calibration curve to calculate brain temperature in eight swine during control, hypothermia, ischemia, postischemia, or death, over a temperature range of 23-40 degrees C. A plot of 61 temperatures determined from 1H NMR versus temperatures measured from an optical fiber probe sensor implanted 1 cm into the cerebral cortex showed excellent linear agreement (slope = 1.00 +/- 0.03, r2 = 0.96). We conclude that 1H NMR spectroscopy presents a practical means of making noninvasive measurements of brain temperature with an accuracy of better than +/- 1 degree C.