The snowflake that appears in videos is studied.The snowflake is randomly distributed in images and it has a certain velocity.The temporal frequencies due to object and camera motions are similar to those produced by snow.A correlation model is used to capture the dynamics of snow.A physics-based motion blur model is used to explain the photometry of snow.Based on these models,we develop an efficient algorithm(DRS) to detect and remove snowflake from videos.The effectiveness of our algorithm is demonstrated by experiments.The algorithm described in this paper can be used in video surveillance.vision based navigation,tracking,license plate recognition and video/movie editing.