Measuring and estimating the traction coefficient is necessary to improve transmitting efficiency and design compact, lightweight toroidal continuously variable transmissions (T-CVTs). However, few attempts have been made to measure and estimate the traction coefficient of T-CVTs under practical usage conditions, and the design of T-CVTs has used extrapolated values from traction coefficients measured under low-power conditions. Therefore, we developed a high-power two-roller traction tester to clarify variation trends in traction curves under operating conditions similar to a T-CVT. The results showed a nearly linear relation between the maximum traction coefficient and the roller surface temperature. Furthermore, the change rate of the maximum traction coefficient with respect to roller surface temperature was dependent on the maximum contact pressure. This paper also compares several traction models under practical operating conditions. A viscoelastoplastic model was constructed and compared with a conventional elastoplastic model. In a wide range of operating conditions, the viscoelastoplastic model showed small differences in the maximum traction coefficient between measured and calculated curves compared with the elastoplastic model. The traction tester and the traction model contribute to building a traction curve database to make T-CVTs more compact, lightweight, and efficient.