A program of development and experimental validation of a multidimensional spray prediction method, based on the discrete droplet model, has been broadened to include computational investigations of the effects of random perturbations of the injection velocity on the spray characteristics, and further detailed examination of the spray structure and development. The results demonstrate strong dependence of the predicted spray penetration length on the precise start-of-injection time and injection velocity data, and relative insensitivity to subsequent variations of the injection velocity. Specifically, it is found that under imposition of random variations of the injection velocity, the variation of the spray-tip penetration and velocity remain smooth, bearing no correspondence to the instantaneous spray injection velocity. The detailed computational investigation of the spray structure and development provides insight into the complex relationship between the penetration length and injection velocity, and highlights the significant influence of spray-front dilution on its detection accuracy, and the discrepancy with predictions at large spray penetration lengths. The novel adoption of a line-of-sight integral-liquid-density criterion for identification of the spray front in the calculations is shown to yield marked improvement of the apparent predictive accuracy.