We present a new model for turbulent Prandtl number that provides an improvement of prediction capabilities of the SST turbulence model in application to wall heat transfer problems. The model was calibrated using Kader’s empirical correlation for near-wall temperature profile. To get an initial assessment of the model we performed computations of the fully developed flow in a round tube and a flat channel with Prandtl number varying from 0.004 to 95; the simulation results were validated against benchmark DNS data and empirical correlations for the Nusselt number. According to the tests, applying the new model resulted in considerable reduction of the Nusselt number prediction error (by factor two and more) in the whole range of Prandtl number considered; the most pronounced effect was observed at Prandtl number values below 0.1.