Multidimensional modeling is one of the chief techniques in conceptual design of data warehouse. If ER models of traditional databases can be directly transformed into multidimensional models, it would efficiently shorten the life circle of data warehouse design. Based on the traditional databases, an algorithm of the transformation from ER models to multidimensional models is presented. Those ER models under some conditions can be transformed, and the resulting multidimensional models conform to the standard of the generalized multidimensional normal form, ensuring the validity of analytical computations on multidimensional databases. Furthermore, the resulting multidimensional models favor an efficient physical database design.