Intrasubject reproducibility of metabolite quantitation in three-dimensional proton magnetic resonance spectroscopic imaging (3D-MRSI) was investigated in 10 healthy volunteers over five separate sessions using two echo times (TEs): 144 and 30 ms. The use of a Gill-Thomas-Cosman (GTC) stereotactic head frame enabled precise subject repositioning and immobilization. Metabolite levels from each voxel in the volume of interest (VOI) were quantified using the Linear Combination of Model spectra (LCModel) analysis algorithm, and coefficients of variation (CVs) were calculated. Standard error estimates (%SD or Cramer-Rao lower bounds) generated by LCModel were used as a confidence filter. The 95% confidence interval (CI) was found for each metabolite, providing an indication of the normal fluctuation expected for 3D-MRSI. In vivo, median CVs at the %SD < or = 20 level were found to be (%CV for TE = 144 and 30 ms, respectively): N-acetyl-aspartate plus N-acetyl-aspartyl-glutamate (NAA): 10.2% and 13.5%; creatine plus phosphocreatine (Cr), 14.4% and 21.7%; and choline-containing compounds (Cho), 15.2% and 18.4%. Relaxing the statistical filtering criteria to %SD < or = 30 increased median CVs by less than 5% and permitted in vivo quantitation reproducibility to be evaluated for glutamine plus glutamate (Glx) and myoinositol (Ins) for TE = 30 ms, yielding CVs of 24.0% and 21.0%, respectively.