This work examines the Subgrid-scale (SGS) model performance and effects of grid resolution for fuel spray simulations within a Lagrangian-Eulerian framework. The widely studied SGS models for large eddy simulation (LES), namely, (1) Smagorinsky model, (2) one-equation dynamic structure model, and (3) no SGS model, are investigated and compared with the Reynolds averaged Navier-Stokes (RANS) approach using the RNG k - epsilon model. The simulation results are also compared against experimental data. For each turbulence model, simulations are performed with different smallest grid sizes ranging from 500 to 31.25 mu m using adaptive mesh refinement with a base grid size of 1 mm. The corresponding smallest grid size to nozzle diameter ratios are 5-0.3. Two diesel surrogates, namely, n-heptane and n-dodecane, are studied under nonreacting conditions relevant for compression ignition engine applications. Experimental data from Sandia National Laboratory through the Engine Combustion Network (ECN) are used for validation purposes. The qualitative comparisons are conducted for the instantaneous mixture fraction and temperature contours. Quantitatively, predicted global spray characteristics of liquid spray and vapor penetration, as well as radial and axial mixture fraction and axial velocity profiles at different locations, are compared against the measurements. Additionally, five different injection realizations are simulated for all the models, with a smallest grid size of 62.5 mu m to capture cycle-to-cycle variations. The required grid resolution for grid convergence for LES spray simulations is discussed based on the experimental comparisons. With the dramatic increase in computational resources in the past decade, this study indicates that LES is a viable alternative to RANS for engine sprays, since it is more predictive in capturing flow structure and local spray characteristics, with reasonable wall-clock times.