The KH-RT (Kelvin-Helholtz-Rayleigh-Taylor) breakup and dynamic structure subgrid-scale (DS SGS) models have been widely utilized to predict the spray evolutions in diesel engines, while the prediction accuracy is highly dependent on the tunable constants. Most of the tuning efforts for both of the above models are limited to the nonevaporating conditions or specified ambient conditions with the Reynolds averaged Navier-Stokes (RANS) approach, and then the tunable constants are supposed to be valid for other conditions. The purpose of this work is to evaluate whether this hypothesis is effective and how large a deviation it will make within the large eddy simulation (LES) framework for the evaporating diesel sprays. Firstly, the spray visualization experiments were conducted under various conditions. Then, the time-efficient Box-Behnken design (BBD) methodology was employed to comprehensively evaluate the sensitivity of prediction accuracy to the tunable constants. The results show that the KH breakup time constant B1 is the most dominant constant for the liquid penetration, while three constants in the DS SGS model significantly affect the vapor penetration. The optimum constants were determined under a specified evaporating condition, which were different from those obtained with RANS. Finally, the performance of the above optimum constants on predicting the spray characteristics was evaluated under further operating conditions. It is found that the discrepancy between the experimental and predicted results varied from 2% to 20% for other conditions, indicating that the optimum constants obtained based on a specified condition cannot provide a satisfactory result, especially when the disturbance intensity and amount of air entrainment into the sprays are changed.