The filtered scalar dissipation rate is an important variable required in many turbulent combustion models used in large eddy simulations (LES). In the present study, direct numerical simulations (DNS) of reacting and non reacting turbulent mixing layers are used to evaluate different models for the filtered scalar dissipation rate. This study is conducted at elevated temperature and pressure conditions relevant to engine applications. The models evaluated are the turbulent diffusivity model, k-epsilon model, the strain rate tensor (SRT) model and the subfilter kinetic energy (SKE) model. It is found that the SRT model is the best choice when considering the performance and ease of implementation in LES codes. DNS results are also used to evaluate the marginal PDF for scalar dissipation rare. It is found that an exponential PDF works well for low values of scalar dissipation rare and smaller filter sizes whereas a lognormal PDF works well for larger values of scalar dissipation rare and larger filter sizes. A model is also formulated for the variance of the scalar dissipation rare to be used when employing the lognormal PDF by relating it to the mean and variance of the mixturc fraction. (C) 2015 Elsevier Ltd. All rights reserved.