In this paper, a new distance for matrix observations called generalized Mahalanobis distance is introduced, some of its properties are studied, and its distribution is obtained for the observations of the matrix variate elliptically contoured distributions. Also, as a significant application, the introduced distance is used in detecting matrix outliers, and its method is described. Finally, some examples are provided for illustrative purposes, and the performance of the presented approach of detecting outliers is investigated by a simulation study.