In C2C e-commerce systems, transactions are anonymous, random and dynamic. Since the transaction information is exchanged between the partners by the virtual network, the partners lack the basic trust foundations and there exist high risks in the process of the transactions. One of the efficient ways to reduce the transaction risk is to evaluate the seller's trustworthiness and help the buyer make scientific decision by trust models. From the buyer's perspective, this paper presents a C2C dynamic trust algorithm (CDTA) for the e-commerce environment. The algorithm takes into account the attributes of trust and trust network, such as the time sensitiveness, the asymmetry of the trust, and the transitivity and selectivity of the trust propagation paths. First, the direct trustworthiness of the buyer to the seller is computed by the transaction experience between them. Second, the reference trustworthiness is computed from the buyer's friends in the trust network according to the recommendation confidence. Finally, the trust of the buyer to the seller is acquired through the integration of the direct trustworthiness and the reference trustworthiness with the trust adjusting factor. The experiments show that the granularity of the trust evaluation is more fine-grained and the evaluation result is more objective than existing work. On the other hand, the similarity review can help the buyer sift out the reference nodes meeting with the buyer's preference, make the reference trustworthiness more credible, and resist the attacks from malicious nodes.