Target selecting is an important aspect of military operational planning. The Markov Decision Process(MDP) method was used to solve the multi-phase target selecting problem which has complex relations among targets. Firstly, the and-or tree was used to describe the relations among the layers of the target system of system(TSoS), and a Discrete Time Markov Decision Process(DTMDP) method was proposed for modeling target selecting whose objective was to neutralize the TSoS. Secondly, an LRTDP algorithm based heuristic was proposed to give the estimate value of the current state of the TSoS, which was calculated by considering the potential resource consumption and failure probability of the evolution process from the current state to the lapse state of the TSoS, and the heuristic can effectively exclude the intermediate states which cannot be transferred to the lapse state, in order to reduce the huge search space of the model because of the complex relations among targets. Finally, a case was proposed to validate the method. The results show that the method is intuitive and practical, and can facilitate the target selecting decision making when there are complex relations among the targets.