High-precision Mg isotopic data are reported for $100 well-characterized samples (granites, loess, shales and upper crustal composites) that were previously used to estimate the upper continental crust composition. Magnesium isotopic compositions display limited variation in eight I-type granites from southeastern Australia (d 26 Mg = À0.25 to À0.15) and in 15 granitoid composites from eastern China (d 26 Mg = À0.35 to À0.16) and do not correlate with SiO 2 contents, indicating the absence of significant Mg isotope fractionation during differentiation of granitic magma. Similarly, the two S-type granites, which rep-resent the two end-members of the S-type granite spectrum from southeastern Australia, have Mg isotopic composition (d 26 Mg = À0.23 and À0.14) within the range of their potential source rocks (d 26 Mg = À0.20 and +0.15) and I-type granites, suggesting that Mg isotope fractionation during crustal anatexis is also insignificant. By contrast, d 26 Mg varies significantly in 19 A-type granites from northeastern China (À0.28 to +0.34) and may reflect source heterogeneity. Compared to I-type and S-type granites, sedimentary rocks have highly heterogeneous and, in most cases, heavier Mg iso-topic compositions, with d 26 Mg ranging from À0.32 to +0.05 in nine loess from New Zealand and the USA, from À0.27 to +0.49 in 20 post-Archean Australian shales (PAAS), and from À0.52 to +0.92 in 20 sedimentary composites from eastern China. With increasing chemical weathering, as measured by the chemical index of alternation (CIA), d 26 Mg values show a larger dispersion in shales than loess. Furthermore, d 26 Mg correlates negatively with d 7 Li in loess. These characteristics sug-gest that chemical weathering significantly fractionates Mg isotopes and plays an important role in producing the highly var-iable Mg isotopic composition of sedimentary rocks.