Biexponential T(2) relaxation of the localized water signal can be used for segmentation of spectroscopic volumes. To assess the specificity of the components an iterative relaxation measurement of the localized water signal (STEAM, 12 echo times, geometric spacing from 30 ms to 2000 ms) was combined with magnetization transfer (MT) saturation (40 single lobe pulses, 12 ms duration, 1440 degrees nominal flip angle, 1 kHz offset, repeated every 30 ms). Voxels including CSF were examined in parietal cortex and periventricular parietal white matter (10 each), as well as 13 voxels in central white matter and 16 T(1)-hypointense non-enhancing multiple sclerosis lesions without CSF inclusion. Biexponential models (excluding myelin water) were fitted to the relaxation data. In periventricular VOIs the component of long T(2) (1736 +/- 168 ms) that is attributed to CSF was not affected by MT. In cortical VOIs this component had markedly shorter T(2)'s (961 +/- 239 ms) and showed both attenuation and prolongation with MT, indicating contributions from tissue. MS lesions and central WM showed a second tissue component of intermediate T(2) (160-410 ms). In white matter similar MT attenuation indicated strong exchange between the two tissue components, prohibiting segmentation. In MS lesions, however, markedly less MT of the intermediate component was found, which is consistent with decreased cellularity and exchange in a region that is large compared to diffusion motion.