Several pioneering studies have demonstrated that localized 31 P NMR spectroscopy of the human heart might become an important diagnostic tool in cardiology. The main limitation is due to the low sensitivity of these experiments, allowing only crude spatial resolution. We have implemented a three-dimensional version of SLOOP ("spectral localization with optimal pointspread function") on a clinical instrument. SLOOP takes advantage of all available a priori information to match the size and the shape of the sensitive volumes to the anatomical structures in the examined subject. Thus, SLOOP reduces the contamination from adjacent organs and improves the sensitivity compared to conventional techniques such as ISIS or chemical shift imaging (CSI). Initial studies were performed on six healthy volunteers at 1.5 T. The good localization properties are demonstrated by the absence of resonances from blood in the heart spectra, and by PCr-free spectra from the liver. Compared to conventional CSI, the signalto-noise ratio of the SLOOP heart spectra was improved by approximately 30%. Taking into account the varying excitation angle in the inhomogeneous B 1 field of the surface coil, the SLOOP model computes the local spin saturation at every point in space. Therefore, no global saturation correction is required in the quantitative evaluation of local spectra. In this study, we found a PCr/␥-ATP ratio in the left ventricular wall of 1.90 ؎ 0.33 (mean ؎ standard deviation).