An oxide film etched in a magnetically-enhanced CHF3/CF4 plasma has been modeled using neural networks. The etch process was characterized with a 2(4-1) factorial experiment. The experimental factors and ranges include 20-80 sccm CHF3 flow rate, 10-40 sccm CF4 flow rate, 300-800 W radio frequency (RF) power, and 50-200 mTorr pressure. Radicals (F, CF1, and CF2) measured with optical emission spectroscopy were related to etch responses. In the experiments conducted, increases in etch rate generally corresponded to decreases in nonuniformity. Etch nonuniformity was strongly dependent on RF power. The relative significance of polymer deposition and ion bombardment was separated and explained. The root-mean-squared prediction errors are 174 Angstrom/min and 0.425% for etch rate and etch nonuniformity models, respectively.