A method for Mg isotope measurement using MC-ICP-MS has been set up. T he authors tried to make a high-precision analysis of magnesium isotopes by MC-ICP-MS for Mg reagent, waters and minerals. The "sample-standard" bracketing technique was used to correct spectrometer mass-dependent fractionation. By arranging narrow entry slit on MC-ICP-MS, the interferences of such molecules as C2+, C2H+, C2H2+, CN+ and NaH+ from Ar gas, air and acid can be decreased to the minimum. Keeping the concentration of samples the same as that of the standard is a good way to obtain high-precision Mg measurement. Through a lot of experiments on matrix element interference, matrix effects on Mg measurement can be realized. Keeping［element］/［Mg］concentration ratio lower than 0.05 can minimize isobaric interference and matrix effects. External repeatability (2SD) of Mg isotope measurement estimated by the authors' laboratory standard CAGS1-Mg relative to CAGS2-Mg is 0.180‰for δ26Mg and 0.090‰ for δ25Mg in a long period. The data in three 25Mg/24Mg versus 26Mg/24Mg plots all fall on the mass-dependent fractionation line with a slope of ～0.5. This means the method put forward by the authors is of high precision and suffers no interference. Relative to DSM3, Mg isotope composition of samples changes at 2.790‰ for δ26Mg and 1.282‰ for δ25Mg. 0.399‰ of δ26Mg from CAGS1-Mg is the maximum, whereas -2.091‰ of δ26Mg from Kanansi lake water is the minimum.