Instrumental mass fractionation (IMF) of isotopic SIMS analyses (Cameca 1280HR, CRPG Nancy) was predicted by response surface methodology (RSM) for 18O/16O determinations of plagioclase, K-feldspar and quartz. The three predictive response surface models combined instrumental and compositional inputs. The instrumental parameters were: (i) X and Y position, (ii) LT1DefX and LT1DefY electrostatic deflectors, (iii) chamber pressure and, (iv) primary-ion beam intensity. The compositional inputs included: (i) anorthite content (An%) for the plagioclase model and, (ii) orthoclase (Or%) and barium (BaO%) contents for the K-feldspar model. The three models reached high predictive powers. The coefficients R2 and prediction-R2 were, respectively, 90.47% and 86.74% for plagioclase, 87.56% and 83.17% for K-feldspar and 94.29% and 91.59% for quartz. The results show that RSM can be confidently applied to IMF prediction in stable isotope SIMS analyses by the use of instrumental and compositional variables.