We have performed a multicentre trial to assess the performance of three techniques for absolute quantification of cerebral metabolites using in vivo proton nuclear magnetic resonance (NMR). The techniques included were 1) an internal water standard method, 2) an external standard method based on phantom replacement, and 3) a more sophisticated method incorporating elements of both the internal and external standard approaches, together with compartmental analysis of brain water. Only the internal water standard technique could be readily implemented at all participating sites and gave acceptable precision and interlaboratory reproducibility. This method was insensitive to many of the experimental factors affecting the performance of the alternative techniques, including effects related to loading, standing waves and B1 inhomogeneities; and practical issues of phantom positioning, user expertise and examination duration. However, the internal water standard method assumes a value for the concentration of NMR-visible water within the spectroscopic volume of interest. In general, it is necessary to modify this assumed concentration on the basis of the grey matter, white matter and cerebrospinal fluid (CSF) content of the volume, and the NMR-visible water content of the grey and white matter fractions. Combining data from 11 sites, the concentrations of the principal NMR-visible metabolites in the brains of healthy subjects (age range 20-35 years) determined using the internal water standard method were (mean+/-SD): [NAA]=10.0+/-3.4 mM (n=53), [tCho]=1.9+/-1.0 mM (n=51), [Cr + PCr]=6.5+/-3.7 mM (n=51). Evidence of system instability and other sources of error at some participating sites reinforces the need for rigorous quality assurance in quantitative spectroscopy.