In this paper, we study a method to translate the style of an input motion which is obtained by a motion capture system. For this aim, we model the difference of motions coming from the actor's attribute, such as gender and age. Especially, we focus on the difference coming from the gender, and model the difference of poses (joint angles) between males and females by using a Gaussian process regression. In training the model, we utilize Multifactor Gaussian Process Model to align the training set of motions in cyclic phase. In the experiments, we train the filter with walking motions of males and females, then we apply the filter to walking, running, and jumping motions of males. The results demonstrate that the filter successfully translates the motion into a womanish one.