We carried out long (145 ms) and short (25 ms) echo time spectroscopic imaging of the brain (chemical-shift imaging, CSI) on two occasions 1 week apart on 15 healthy individuals. We found coefficients of variation (CVs) generally in the range 10-25% for long and 15-30% for short echo-time measurements. The CVs of metabolite ratios were higher by about 5-10%. Limits of agreement (defined as mean+/-2 SD of the week 1-week 2 differences) were wider at the shorter echo time. The modest repeatability may be due in part to the difficulty of repositioning spectroscopic voxels at a scale of 1 mm. The generally higher CVs and wider limits of agreement at TE25 ms suggest that the increased spectral complexity more than offsets the theoretical advantage of increased signal at short echo-times. Analysis of variance general linear modelling of metabolites and metabolite ratios showed that, in general, the subject, region of the brain and hemisphere were more important than the occasion in explaining the variability of results. Unless information on short-T2 metabolites is specifically required, better results can probably be achieved with longer echo-times. The magnitude of the CVs needs to be taken into account in the calculation of sample size for cross-sectional or linear studies.