This report proposes a method of improving the temperature prediction model for traction drive contact portion in order to improve prediction accuracy of the maximum traction coefficient, and then describes verification of this method.In our previous report, a method of estimating the maximum traction coefficient by expressing conditions inside the contact ellipse using a simple combination of viscosity and plasticity was proposed. For the rise in oil film temperature, a calculation model is used that considers maximum temperature to be the typical value. Furthermore, a thin film temperature sensor technology was developed to directly measure the temperature of traction contact of a four-roller experimental apparatus and a variator in an actual transmission, and its validity was confirmed. When measured values of the traction coefficient μ were compared with calculated ones, it was learned that μ estimation accuracy decreased under certain low-contact-pressure and high-sliding conditions. The factor that caused μ estimation accuracy to decrease under these load conditions was presumed to be an error in oil film temperature estimation. Therefore, by partially improving the temperature prediction model, that factor was identified and examined.In the end, the maximum traction coefficient was estimated using the new calculation model and compared with the measured values.