The execution of business process usually requires the collaboration of many staff members. When the capability of a staff member is determined, the collaboration becomes the dominating influence to the performance of process instances. In general, the better collaboration of the actors who perform collaborative tasks, the higher performance the workflow instance would achieve. In this paper, an approach to optimize staff assignment is proposed based on the collaboration patterns with high performance. Firstly, it computes the compatibility of actors when performing activities based on the historical logs. Afterwards, it mines collaboration patterns with high compatibility from process logs and matches them with the process for staff assignment by means of encoding. Experimental results demonstrate this approach can assign staff in a workflow for maximum collaboration efficiently.