A piloted turbulent natural-gas diffusion flame is investigated numerically using a 2D elliptic Monte Carlo algorithm to solve for the joint probability density function (PDF) of velocity and composition. Results from simulations are compared to detailed experimental data: measurements of temperature statistics, data on mean velocity and turbulence characteristics and data on OH. Conserved-scalar/constrained-equilibrium chemistry calculations were performed using three different models for scalar micro-mixing: the interaction by exchange with the mean (IEM) model, a coalescence/dispersion (C/D) model and a mapping closure model. All three models yield good agreement with the experimental data for the mean temperature. Temperature standard deviation and PDF shapes are generally predicted well by the C/D and mapping closure models, whereas the IEM model gives qualitatively incorrect results in parts of the domain. It is concluded that the choice of micro-mixing model can have a strong influence on the quality of the predictions. The same flame was also simulated using reduced chemical kinetics obtained from the intrinsic low-dimensional manifold (ILDM) approach. Comparison with the constrained-equilibrium results shows that the shape of the OH concentration profiles is recovered better in the ILDM simulation, and that the ILDM reduced chemical kinetics can correctly predict super-equilibrium OH.