A direct numerical simulation (DNS) database is used to develop a model of subgrid-scale temperature fluctuations for use in large-eddy simulations of turbulent, reacting hypersonic flows. The proposed model uses a probability density representation of the temperature fluctuations. The DNS database reveals a physically consistent relation between the resolved-scale flow conditions that may be used to predict the standard deviation of the Gaussian probability density function (PDF). The model is calibrated and tested by comparison to simulations of decaying isotropic turbulence. The conditional single-variable PDF model is found to capture the fluctua-tions in temperature and product formation.