Zircon stability in silicate melts—which can be quantitatively constrained by laboratory measurements of zircon saturation—is important for understanding the evolution of magma. Although the original zircon saturation model proposed by Watson and Harrison (Earth Planet Sci Lett 64(2):295–304, 1983) is widely cited and has been updated recently, the three main models currently in use may generate large uncertainties due to extrapolation beyond their respective calibrated ranges. This paper reviews and updates zircon saturation models developed with temperature and compositional parameters. All available data on zircon saturation ranging in composition from mafic to silicic (and/or peralkaline to peraluminous) at temperatures from 750 to 1400 °C were collected to develop two refined models (1 and 2) that may be applied to the wider range of compositions. Model 1 is given by lnC_Zr(melt) = (14.297 ± 0.308) + (0.964 ± 0.066)·M − (11113 ± 374)/T, and model 2 given by lnC_Zr(melt) = (18.99 ± 0.423) − (1.069 ± 0.102)·lnG − (12288 ± 593)/T, where C_Zr(melt) is the Zr concentration of the melt in ppm and parameters M [= (Na + K + 2Ca)/(Al·Si)] (cation ratios) and G [= (3·Al₂O₃ + SiO₂)/(Na₂O + K₂O + CaO + MgO + FeO)] (molar proportions) represent the melt composition. The errors are at one sigma, and T is the temperature in Kelvin. Before applying these models to natural rocks, it is necessary to ensure that the zircon used to date is crystallized from the host magmatic rock. Assessment of the application of both new and old models to natural rocks suggests that model 1 may be the best for magmatic temperature estimates of metaluminous to peraluminous rocks and that model 2 may be the best for estimating magmatic temperatures of alkaline to peralkaline rocks.