An improved approach is presented for the hybrid Eulerian-Lagrangian modeling of turbulent two-phase flows. The hybrid model consists of a nonlinear k - epsilon model for the fluid flow and an efficient Lagrangian trajectory model for the particulate flow. The improved approach avoids an empirical correlation required to determine the dispersion width for the existing Stochastic-Probabilistic Efficiency Enhanced Dispersion (SPEED) model. The improved SPEED model is validated using experimental data for a poly-dispersed water spray interacting with a turbulent annular air jet behind a bluff-body. Numerical results for the number-mean and Sauter-mean droplet diameters, as well as mean and fluctuating droplet velocities are compared with the experimental data and with the predictions of other dispersion models. It is demonstrated that higher computational efficiency and smoother profiles of Sauter-mean diameter can be obtained with the improved stochastic-probabilistic model than with the eddy-interaction model.