To solve the security issues in a mobile agent system, the access control is always employed. Existing access control models enable the agent's identity, trust and other information to influence the security decisions. We argue that the path, which a mobile agent travels along, should be taken into consideration and affect what the agent can do dynamically. An access control model is proposed in this paper which captures the path history of the agent and reaches a decision according to both whether the path matches a path pattern set and whether the path can follow a host patch set. The flexibility of this model is shown and some issues in implementing such a model are discussed as well.