Metaphor computation is an important problem of natural language processing. In this work, an intensive study of metaphor identification is carried out from different angles, including linguistical, psychological, and cognitive angles. Human categorization is a dynamic process of measuring difference between objects from different angles. Therefore, an idea about dynamic categorization is proposed from multiple angles to recognize metaphors, which is different from traditional metaphor recognition methods. The research involves three aspects:how to get features of concepts, how to choose angles by features, and how to measure difference based on a specific angle. Then, an experiment of nominal metaphor recognition is performed based on dynamic categorization. The experimental results show that the accuracy of metaphorical/literal references recognition can reach 85.4%. It supports the validity, efficiency of the proposed method.