Longitudinal relaxation of brain water (1)H magnetization in mammalian brain in vivo is typically analyzed on a per-voxel basis using a monoexponential model, thereby assigning a single relaxation time constant to all (1)H magnetization within a given voxel. This approach was tested by obtaining inversion recovery (IR) data from gray matter of rats at 64 exponentially spaced recovery times. Using Bayesian probability for model selection, brain water data were best represented by a biexponential function characterized by fast and slow relaxation components. At 4.7T, the amplitude fraction of the rapidly relaxing component is 3.4% +/- 0.7% with a rate constant of 44 +/- 12 s(-1) (mean +/- SD; 174 voxels from four rats). The rate constant of the slow relaxing component is 0.66 +/- 0.04 s(-1). At 11.7T, the corresponding values are 6.9% +/- 0.9%, 19 +/- 5 s(-1), and 0.48 +/- 0.02 s(-1) (151 voxels from four rats). Several putative mechanisms for biexponential relaxation behavior were evaluated, and magnetization transfer (MT) between bulk water protons and nonaqueous protons was determined to be the source of biexponential longitudinal relaxation. MR methods requiring accurate quantification of longitudinal relaxation may need to take this effect explicitly into account.