Overweight and obesity are prevalent nutritional disorders that are becoming a worldwide epidemic (1). They are associated with an increase in total body fat, elevated blood pressure, insulin resistance, increased incidence of type 2 diabetes, etc. It is well known that in addition to over-accumulation of subcutaneous and visceral adipose tissue, extra-(EMCL) and intramyocellular (IMCL) lipids in skeletal muscle are important factors in the pathogenesis of these concomitant diseases (2,3). IMCL stores have been found to be a potent marker of insulin resistance in obese (4,5) and non-obese adults (6), as well as in nondiabetic offspring of type 2 diabetic subjects (7,8).Most previous attempts to quantify IMCL were based on the analysis of muscle biopsy samples (9 -11). However, measures of IMCL derived in this manner lack sufficient accuracy and sensitivity to distinguish between IMCL and EMCL content. Moreover, the samples in the above-cited studies were small, and it is technically difficult to assess fat biochemically (12). It was recently demonstrated that proton magnetic resonance spectroscopy ( 1 H MRS) is a convenient noninvasive technique that allows the discrimination and measurement of IMCL and EMCL pools in vivo (13,14). 1 H MRS has been used to measure the IMCL level, in most cases with single-voxel techniques (5)(6)(7)(8)(12)(13)(14) and a voxel volume greater than 1 cm 3 . In obese subjects, such large voxels often contain regions with increased EMCL, which contaminate the IMCL spectrum due to severe signal overlap. Single-voxel spectroscopy (SVS) is especially susceptible to this problem because postacquisition voxel repositioning is impossible and the selected voxel location may not yield spectra with acceptable IMCL and EMCL separation. Moreover, the problem of voxel location complicates inhomogeneous distribution of IMCL and EMCL within the same muscle and between muscles (15-17). This problem is often underestimated in studies that seek to measure IMCL in a predefined region (e.g., for correlations with other methods), and in longitudinal studies. To overcome these difficulties, MR spectroscopic imaging (MRSI) techniques with a measured matrix of up to 32 ϫ 32 (phase-encoding steps) have been used (15,16,18). Despite the smaller size of the voxels (0.25-0.5 cm 3 ), there are still severe overlaps of the IMCL and EMCL spectral components in many voxels. In addition, MRSI spectra are contaminated by a signal bleeding from bone marrow and subcutaneous and interstitial fat as a consequence of limited k-space sampling. Several methods have been reported to suppress fat signal bleeding with postacquisition processing (16,19), but none of these methods offers a definite solution. Standard SI with a large number of phase-encoding steps (e.g., 128 ϫ 128 or more) can be efficient in removing the signal bleeding. However, it requires unacceptably long acquisition times to acquire all k-space encodings.The purpose of the present study was to investigate whether high-resolution SI with a large number of pha...