Although the Eulerian particle flamelet model (EPFM) recently proposed by Barths et al. [Proc. Combust. Inst. 27 (1998) 1841-1847] has shown the potential capabilities to realistically predict detailed pollutant (NOx, soot) formation in a turbulent reacting flow occurring within practical combustion devices, there still exists room to improve the predicative capability in terms of local flame structure and turbulence-chemistry interaction. In this study, the EPFM approach was applied to simulate two turbulent nonpremixed jet flames of CO/H-2/N-2 fuel having the same jet Reynolds number but different nozzle diameters, and the capability of predicting the NOx formation as well as both similarity of major species and sensitivity of minor species to fluid-dynamic scaling for the two flames has been assessed deeply in terms of both conditional and unconditional mean structures. The present results indicate that the original EPFM substantially overpredicts the conditional scalar dissipation rate at the downstream region and consequently underpredicts the streamwise decay of superequilibrium radical concentrations to the equilibrium state. In this study, in order to correctly estimate the averaged conditional scalar dissipation rate, a new modeling of the conditional scalar dissipation rate based on a least-squares fit through a mass weighted spatial distribution has been devised. In terms of both conditional and unconditional means, the EPFM utilizing this new procedure yields nearly the same results as the Lagrangian flamelet model, and provides closer agreement with experimental data than the original EPFM approach. (C) 2008 The Combustion Institute. Published by Elsevier Inc. All rights reserved.