We report a result of an experimental study on properties of pose representations for 3 DOF linear pose estimations with 100 CG objects. We use linear regression as a pose estimation method. First, we explain a method of linear pose estimation and two properties of pose representations. Next, we use four pose representations (rotation matrix, ZYX Eulerangle, exponential map, and unit quaternions), for pose estimation experiments, and compareestimation errors. We show that estimation error of rotation matrix is significantly smaller than other representations by using pairwise t–test.