Workflow technology is an enabling technology for enterprises to realize business process automation. It coordinates activites, data and people. Staff assignment is an important part of coordination between activites and people. The objective of staff assignment is to ensure correct assignment of activites to people.Traditional staff assignment approaches deal with policies and algorithms based on the matching knowledge between people and activities, from the viewpoint of operation and optimization. However, in the context of workflow, it is very difficult to obtain such matching knowledge. Besides, the people involved are highly dynamic. Therefore, this thesis tries to study workflow runtime staff assignment through history information left by people who executed activities. The purpose of this thesis is to solve the problem of staff assignment under the condition of incomplete knowledge, and tries to discover the characteristics between activities,Major contributions of this thesis are as follows:1. A machine learning based semi-automatic staff assignment approach is proposed to meet the requirement of dynamic staff assignment during workflow runtime for some critical activities. This approach learns the staff assignment patterns from the event log of activity instances, and recommends suitable candidate actors for those critical activities to process instance monitors.2. Based on the "pull" behavior of actors during workflow runtime, a temporal interval logic based associated work items mining approach is proposed. This approach tries to discover parallel associated work items by analyzing the start and end time of work items, i.e. temporal logic information, and recommends these work items to actors.3. A decision tree based activity execution time prediction approach is proposed. The execution time of work items is relevant to the context (e.g. actors, difficulty of cases etc.). By analyzing the event log of previous activities in workflow event log, this approach tries to predict current activity\'s execution time. Finally, in order to validate the aforementioned approaches, experiments have been performed on real-world workflow event log of three domestic manufacturing enterprises. In addition, a design of workflow system with runtime staff assignment is given and we have realized part of the design in the TiPLM system developed by National Engineering Research Center of Supporting Software for Enterprise Information Technology.