The emerging high efficiency video coding (HEVC) standard uses residua l quadratic tree (RQT) to transform and quantize the predicted residuals after motion estimation (ME).Rate-distortion costs of all the nodes at different RQT depths are compared transversally to sel ect the optimal one or we call it the optimal transform unit (TU).The coding efficiency is improved si gnificantly by the new transform technology.However,huge computation complexity is introduced compare d with previous standard.To solve the above problem,a rapid TU mode decision algorithm is propos ed in this paper focusing on adaptive depth prediction.First of all,weights of temporal-spatia l adjacent CTUs combined with characteristics of RQT depth distribution themselves are used to p redict RQT depth of certain CTU/CUs.The RQT depths of rest CTU/CUs are predicted via temporal-spati al correlation between RQT maximum depths.As a result,unnecessary RQT depth can be skipped or terminated. According to simulation results,conclusions are obtained that the TU computatio n complexity can be reduced by about 50% under standard coding configuration,while maintaining almos t the same RD performance as original standard.Our algorithm reduces TU computation complexit y greatly as expected.Meanwhile,the proposed algorithm effectively addresses the low effici ency problem under complex sequences met by other related algorithms.