Abstract: The traditional text-based search method for CAD models often obtain searched results that may not satisfy with user's design intent.To solve this problem,a new method is proposed to improve the precision of the text-based search by automatically assigning semantic tags to CAD models.First a method based on attributed graph for determining the shape similarity of the CAD models is introduced.A sequence of graph nodes is created based on the adjacent matrix of graph and attributes of nodes.The maximum common graph representing the shape similarity of models is calculated by dynamic programming with the sequence of the graph nodes.Then probability method is used to tag the CAD models automatically according to the similarity between the models.The experiments show that the text-based search for CAD models with semantic tags could enable the users to more accurately locate the models with similar shapes and the user's design intents can be satisfied.