Direct numerical simulation (DNS) and large eddy simulation (LES) of two-dimensional nonpremixed methane jet flames are conducted to assess the performance of subgrid-scale LES models and reduced kinetics mechanisms in transitional and turbulent flows. The LES is via the recently developed "filtered mass density function" (FMDF) method of Jaberi et al. [1]. The FMDF represents the single-point joint probability density function (PDF) of the mass weighted subgrid-scale scalar quantities, and is obtained by solving its transport equation via a Lagrangian Monte Carlo scheme. In the FMDF transport equation, the effects of chemistry appear in a closed form, allowing reliable LES of turbulent flames with complex chemistry models. The LES/FMDF results are appraised by detailed comparisons with DNS data for various reduced and skeletal mechanisms. It is shown that the filtered values of the major and minor species and the compositional structure of the methane flames are accurately predicted by FMDF for all the tested chemistry models. However, the DNS and LES results as obtained by the reduced mechanisms are found to be substantially different than those calculated by the skeletal mechanism in some how conditions. This is consistent with our laminar coflow and counterflow jet results, and indicates the importance of kinetics models in the numerical simulation of transitional/turbulent hydrocarbon flames.