Active appearance model (AAM) is a powerful method for objects feature localization and successfully used in computer vision. Although widely employed, AAMs suffer from a few drawbacks, such as accuracy on the contour of face. We present an Active Appearance Models with contour constraints method from depth data acquired by consumer depth cameras, which could capture video data in 3D, include depth data and color image, under any ambient light conditions based on invisible infrared projection. The facial contour points are acquired from depth data that can be used to restrain the fitting result of AAM and improve the accuracy of the contour part of the result shape by minimize the Euclidean distance in parameter space from the previous state to the solution. Experimental results show the improvement on the facial feature locating.