A novel gas atomization process (GAP) was designed for mass producing high-quality spherical powders from molten polymers. The dependence of the powder characteristics, such as particle size and size distribution on the GAP control variables, and physical properties of the polymers were determined experimentally. Statistical and computational neural network analyses were applied to the data in both the forward and reverse modes, leading to models for predicting the particle size distribution from the process variables and vice versa. The computational neural network (CNN) model predicts an increase in the weight fraction of the smallest particle size range (0-53 mu m) by increasing the polymer melt temperature while decreasing the polymer melt stream size for the more crystalline polyethylene-based material studied. Comparisons of the CNN model predictions with experimental data yields good agreement and demonstrates that the CNN can be used to qualitatively design new experiments for the production of spherical powders from molten polymers, thereby reducing the amount of experimental work required to optimize the process.