For the mobile-agent-based e-commerce environment, most reputation-based trust algorithms are onedimensional. They are only based on the node’s historical transactions, and the services are not taken into account, so the evaluations are coarse-gained. This paper proposes a reputation-based multi-dimensional trust (RMDT) algorithm which makes use of a self-confident coefficient to synthesize the directed and the reference trustworthiness to evaluate the node in the network. The time sensitive function in this paper not only serves as a reward and punishment mechanism, but is also dynamically attenuate in its trustworthiness. Both the transaction evaluation system and the weight system are introduced in the multi-dimensional trust mechanism. RMDT can uncover the influence on trust computation caused by the subjective factors, such as individual predilection and risk attitude. In addition, the sensitivity of RMDT on the single attribute is greatly improved.