The heterogeneous proton MR relaxation decay process in human brain has been investigated by performing region-of-interest and pixel-by-pixel calculations on the multiecho MR images with different repetition times (TR) of human brains using a clinical 1.5-T whole-body superconducting MR scanner. Based on the monoexponential, biexponential, and continuous gaussian distribution relaxation models, first-order proton relaxation parameters (proton density, T 1 and T 2) and higher-order transverse proton relaxation parameters ( T 2-long, T 2-short, T 2-long fraction, T 2-average, and T 2-distribution width) were calculated. On the basis of an F test ( p < .01), the statistical significance of the higher-order (biexponential and distribution) fits over the monoexponential fit was evaluated. Here, a significant improvement in the biexponential fit was found for some of the regions containing the ventricular cerebrospinal fluid (CSF) ( T 2-long = 2780 ± 570 ms; T 2-short = 159 ± 42 ms; T 2-long fraction = 0.51 ± 0.08 ms) due to the partial volume effect but not for most of the white matter (WM). On the other hand, an improvement of fit to WM was obtained when distribution ( T 2-average = 80 ± 8 ms; T 2-distribution half-width = 21 ± 4 ms) as opposed to monoexponential ( T 2 = 89 ± 10 ms) fit was used. As internal controls, tubes of CuSO 4 solution ( T 2 = 1293 ± 128 ms) and agarose gel ( T 2 = 111 ± 10 ms) which have similar T 2 values as the CSF and WM of the brain, respectively, were attached to the human head and imaged concomitantly. No significance improvements in either the biexponential or distribution fits over the monoexponential fit were found for all the controls. In addition to the first-order and higher-order relaxation parameter maps, the monoexponential chisquares, as well as the chisquares ratio (chisquares of the monoexponential fit divided by that of the higher-order fit), maps were also generated. Unlike the higher-order T 2-relaxation parameter maps, the chisquares parameter maps required no selection of any predetermined statistical confidence level. Therefore, these chisquares parameter maps provided a somewhat nonsubjective spatial prolife of the heterogeneous transverse relaxation process in the brain. Our results led us to propose that the use of chisquares parameter maps, together with the first-and higher-order relaxation parameter maps, may further improve the in vivo tissue characterization capability of MRI in future clinical diagnosis and staging of intracranial diseases.