This paper summarizes our results on gearbox failure prediction using infrared camera. Experimental data was taken at the Mechanical Diagnostic Test Bed (MDTB) of the Penn State University. It was observed that there is thermal growth before a gearbox failure occurs. One explanation is that friction between gears increases before the gear breaks. A neural net based image-processing tool was developed by Intelligent Automation, Inc. (IAI) to process the infrared images in real-time. Results show that our tool can detect unusual thermal growth five hours before one of the gear teeth was broken.