Abstract: Large eddy simulations (LESs) of Sandia turbulent non-premixed flame D have been carried out using two sub-grid scale combustion models. One is tabulated chemistry method coupled with three presumed probability density functions (PPDFs), the other is sparse Lagrangian filtered density function (FDF) method. The goal is to show the differences of the two combustion models and highlight the characteristics of turbulent non-premixed flame and its pollutant. Comparisons of statistical results indicate that the distributions of temperature and major species are similar, while CO can be better captured by the FDF model. Interactions between turbulence and flame (TFIs) can be described reasonably by all the three PPDFs, and the difference between them is mainly reflected in the predicted distributions of NO. NO mass fraction is largely overpredicted with Dirac function, and underestimted with Top-hat function. Beta function presents the best performance in the prediction of NO. Regions of high temperature and mass fraction of NO are mainly located around φ=1.0 and rich side. The mass fraction of NO peaks mainly in the region where the scalar dissipation rate of mixture fraction is relatively small. The mass fraction of NO is always highly correlated with local temperature, in the presence of the high temperature piloted flame. It is also highly correlated with the mass fraction of O₂ and H₂O at different axial locations.