In order to reduce the high efficiency video coding (HEVC) inter-fram e prediction complexity,in this paper we propose an HEVC inter pred iction mode selection algorithm based on motion features.First,we use the time- domain co rrelation between adjacent frames in digital video to determine the optimal c odi ng depth through calculating the frame difference dispersion (FDD) of each coding unit (CU) and its four sub-blocks,then we can divid e the CU into three kinds of motion areas according to the regional motion feature (RMF) in the optimal codin g depth,finally we can obtain the candidate inter prediction modes and reduce th e u nnecessary inter prediction process.The experimental results show that this a lgorithm can improve the coding efficiency under the premise of a little coding pe rformance loss.Under two coding structures of low-delay and random-access,com pared with the standard method,our algorithm increases bit rate by 0.89% and 0. 89% and saves coding time by 51.6% and 48.5%,respectively with the same object ive quality.