Gear mechanisms are key components for rotating machinery ranging from automotive, hydraulic systems to aviation systems. As a more reliable, safer, economical fault diagnostic method, vibration and acoustic signatures of such systems have been widely studied. There are only a few numbers of studies incorporating sound and vibration monitoring together, for different working hours of the mechanism, rotating at different operational parameters. A bevel gear test setup was developed in-house to observe the effect of different operating conditions as shaft loading, shaft speed, lubrication level and abrasive contamination along with different operating hours. The system operating condition was also monitored, by obtaining visual photographs of gear teeth. Vibration and sound signals were recorded followed by fast Fourier Transform and Power Spectrum Density computations to extract the features used in developing a Multi-Layer Perceptron (MLP) based Neural Network and a Neural Pattern Recognition algorithm for fault classification purposes. It has been shown that sound and vibration measurements can be confidently used to predict bevel gear fault conditions.