Summary form only given. Bistatic MIMO radar has become an active research area in recent years. In this paper, linear prediction combining with eigenvalue decomposition is applied to DOD and DOA estimation of bistatic MIMO radar. Firstly, according to signal model, the output signals of receivers are depicted by transceiver combing steering vector which is the kronecker product of receiving steering vector and transmitting steering vector of targets. Secondly, in order to estimate DOD and DOA of multiple targets at the same time, useful signals space and noise space are segregated by eigenvalue decomposition of received signals correlation matrix, which also helps to improve the performance of linear prediction in bistatic MIMO scenario with uniform linear arrays. Thirdly, received signals correlation matrix is replaced by noise correlation matrix to cut down computation complexity. Finally, simulation results verify the effectiveness of LP_EVD for multiple targets location of bistatic MIMO radar and the performance of this method is compared with existing methods.