Aiming at the fact of reference dependence and loss averse of decision makers in the actual decision making process, a grey multi-attribute group decision making method of considering the psychological behavior of decision makers is proposed. Firstly, to fully reflect the idea of“rewarding the good and punishing the bad”, the original decision information is normalized by using the“rewarding the good and punishing the bad”transformation operator. Then, the global priority of alternatives is obtained by calculating the priority of benefit-loss value of each alternative relative to other alternatives about attributes under decision makers. Furthermore, to calculate the comprehensive value of alternatives and rank them, a weight model of decision makers is constructed according to the maximum entropy and group's consistency. Finally, a case analysis is given to show the feasibility, rationality and practicability of the proposed method.