The reproducibility of the phosphocreatine to adenosine triphosphate ratio (PCr/ATP) was assessed from cardiac phosphorus-31 (31P) NMR spectra of the human left ventricle acquired with three different localization techniques. Cardiac 31P-NMR spectra (n = 68) were obtained at rest from 16 healthy subjects with three-dimensional (3D) image selected in vivo spectroscopy (ISIS), 1D spectroscopic imaging (SI), or with a combination of 2D ISIS and the 1D SI technique (ISIS + SI). The average PCr/ATP ratios were 1.41 +/- 0.20 for ISIS + SI and 1.31 +/- 0.19 for ISIS and were in the lower range of values obtained in previous studies, mainly because of a lower saturation correction factor for the cardiac PCr/ATP ratio. The SI experiment yielded an average PCr/ATP value of 0.98 +/- 0.20, significantly lower as compared to the correct values obtained with ISIS + SI and ISIS (p < 0.001), underscoring the need for 3D localization to avoid contamination of the NMR signal by liver tissue. Intersubject standard deviations of the PCr/ATP ratio were comparable to values reported previously. For all three localization techniques the absolute intra-examination differences in PCr/ATP (0.06 for ISIS to 0.15 for ISIS + SI) were significantly smaller (p approximately 0.03) than inter-examination differences (0.24 for ISIS to 0.29 for ISIS + SI). Therefore, consecutive acquisition of cardiac 31P-NMR spectra from the same patient during a single examination, e.g. under various cardiac loading conditions, appears to be a reliable approach for metabolic evaluation of heart disease.