Using discretization technique is one of the challenges associated with numerical modeling. In the Large Eddy Simulation (LES) method, this issue becomes more critical because it affects the sub-grid scale (SGS) model and creates errors. Using an LES method, we investigated the numerical errors and compatibility of different discretization methods with SGS in pool fire modeling. First-order, central, and second-order upwind and linear upwind stabilized transport (LUST) schemes were examined in Smagorinsky (S_SGS) and k-equation (K_SGS) sub-grid scales. First-order upwind and second-order upwind methods estimated the velocity field, temperature, and perturbation with a significant error, so that in the plume area, there was 90% error with the first-order upwind and 50% error with the second-order upwind concerning the results of the mean square horizontal velocity perturbation. With an error of 20%, the LUST method had a better agreement with the experimental results. Besides, SGS did not have a significant effect on the results of this method. However, the second-order upwind method was more consistent with the S_SGS model. To improve the results of upwind methods, Quadratic Upstream Interpolation for Convective Kinematics and Monotonic Upwind Scheme for Conservation Laws discretization methods are recommended.