A new algorithm for dimension variation of feature-based models is developed.The algorithm is based on B-Rep/CSG hybrid scheme and operates directly on B-Rep.Product information (including the features locating dimensions and other data for manufacture) will not lose after model variation and modification.Furthermore, the definition and solution of features constraints are also supported.The scheme of directly operating on B-Rep overcomes many drawbacks of other proposed methods, most of which need to redo all of the previous work to implement dimension variation, so the computational cost is expensive and product information will lose.What is presented in this paper makes a new way in the area of parametric design.