Because the number of tasks to consider in user interface (UI) design and evaluation increases, it becomes important to select a set of representative tasks. This thesis proposes a method for identifying the representative tasks based on the task properties and the interaction procedures. The former is identified through hierarchical task analysis (HTA) and represented in a declarative form while the latter is described using the OCD (Operation and Control Diagram) model. OCD is able to abstract interaction knowledge to match the user`s natural thinking. The method performs task clustering based on the procedural similarities between tasks, which are measured by comparing OCDs and applying the method of error-correcting graph matching. A*algorithm is adopted to optimize the matching between two diagrams. The HTA also yields the task property similarity between two tasks. The overall similarity between tasks then is determined by adding two types of similarities. Two experiments were conducted to verify the reasonableness of the knowledge representation and the clustering method, respectively.