A transmission gear failure affects function and performance of the entire machine. Meanwhile, gears may break before expected service life even if bending strength, surface durability and scuffing strength, which are evaluated in the desigh process of gears, are fulfilled. So it is extremely important to establish a gear failure prognosis technology. The existing gear failure prognosis method through vibration, acoustic and temperature measurement can't be universally applied because it can only catch limited sort of signal. Our experiments with case-carburized gears were conducted for the purpose of establishing a practical method. Gear fault progressed from micro pitting to flaking and finally to tooth crack. During the test, gear condition was observed and vibration data was recorded in detail. In this study, Mahalanobis distance was calculated from the vibration data in order to evaluate the gear condition. The result shows that Mahalanobis distance could be an indicator which shows gear condition quantitatively. There is a possibility that a vibration analysis by Mahalanobis distance could be a practical method to prognose gear fault before machine break down.