A new model describing zircon saturation in silicate melts is presented that combines the results of 196 data from new experiments with data from previous experimental studies. In the new experiments, the concentration of Zr in melts coexisting with zircon was determined at temperatures between 800 and 1500 °C for 21 compositions (with alumina saturation index, ASI, from 0.20 to 1.15), containing ~ 1 to 16 wt % FeOT and, for a subset of these conditions, at variable pressure (0.0001 to 4.0 GPa) and water content (0 to 15 wt %). The collated dataset contains 626 data, with 430 from 26 literature studies, and covers conditions from 750 to 1620 °C, (including 45 new data and 106 literature data for temperatures < 1000 °C), ASI 0.20 to 2.00, 0.0001 to 4.0 GPa and 0 to 17 wt % H2O. A limitation of previous models of zircon saturation is the choice of parameter used to describe the silicate melt, which may not be appropriate for all compositions and can result in differences in predicted temperatures of over 200 °C for granitic systems. Here we use optical basicity (Λ), which can be easily calculated from the major oxide components of a melt, to parameterise the composition. Using a non-linear least-squares multiple regression, the new zircon saturation model is:$$\log{\mathrm{Zr}} = 0.96(5) - 5790(95)/T - 1.28(8)P + 12.39(35){\varLambda} + 0.83(9)x.\text{H}{_2}\text{O} + 2.06(16)P{\varLambda} $$
log
Zr
=
0.96
(
5
)
-
5790
(
95
)
/
T
-
1.28
(
8
)
P
+
12.39
(
35
)
Λ
+
0.83
(
9
)
x
.
H
2
O
+
2.06
(
16
)
P
Λ
where Zr is in ppm, T is temperature in K, P is pressure in GPa, Λ is the optical basicity of the melt, x.H2O is the mole fraction of water in the melt, and the errors are 1σ. This model confirms that temperature and melt composition are the dominant controls on zircon solubility. In addition, pressure and melt water content exert small but resolvable effects on the solubility and are included, for the first time, in a model. Using this new calibration, 92% of the predicted temperatures are within 10% of the experimental temperatures for the collated dataset (with an average temperature difference of 57 °C), while predicted temperatures for only 78 and 62% of the collated dataset are within 10% of the experimental temperature (with average temperature differences > 80 °C) using the widely cited Watson and Harrison (Earth Planet Sci Lett 64:295–304, 1983) and Boehnke et al. (Chem Geol 351:324–334, 2013) models, respectively. This new model can be extrapolated to temperatures below those included in the calibration with greater accuracy and when applied to melt inclusions from the Bishop Tuff, gives temperatures that are in excellent agreement with independent estimates.