Detailed electron-microprobe line profiles and small-area compositional maps of zoned garnets in a sample of high-pressure-high-temperature granulite show features inconsistent with commonly applied diffusion models. Larger grains of an early garnet generation have their highest Ca contents in domains away from the rim or inclusions but show a sharp fall in Ca balanced by increased Mg and Fe (and slightly higher X Mg ) towards inclusions and the rim. In domains with secondary biotite, the sharp decrease in Ca is accompanied by variations in X Mg dependent upon proximity to biotite, thus producing one-sided, asymmetric profiles with X Mg lower against biotite. As a consequence, rim compositions of the same grain are different on the sides adjacent and away from biotite and there is no relationship between grain size and rim X Mg . Such a zoning pattern requires that grain-boundary diffusion is as slow as volume diffusion and implies the absence of a diffusion-enhancing grain-boundary fluid phase during the majority of the rock's high-temperature exhumation history. Diffusion models ignoring this probability could yield either cooling rates that were too fast, or extrapolated ages based on closure temperature models that were too old. A second garnet generation in the same rock, grown in a Ca-rich domain resulting from kyanite breakdown, has irregularly distributed patches, identified by compositional mapping, containing higher Ca than the first-formed garnet but at lower X Mg . Use of such garnet compositions for geothermobarometrical determination of the high-pressure granulite stage would clearly lead to erroneous results. The presence of such contrasting garnet compositions in a granulite-facies rock is clearly evidence of disequilibrium, and further supports the proposition that there was a lack of an effective transport medium even at the mm scale.