Abstract: A new reduction method for solving large generalized eigenvalue problems is presented, which can Not only overcome difficulties of computer capacity and give the required accuracy but also save much computer run time.The "dynamic transformation" method suggested by Kuhar is an improvement on Guyan's reduction method. But this method is Now only applied to the eigenvalue problems deduced from modal synthesis techniques and its general mathematical consideration has Not appeared yet.This paper is devoted to the mathematical discussion of Kuhar's idea and deals with general generalized eigenvalue problems. Furthermore, it proposes a new reduction method, called "dynamic reducing iterative method", by virtue of which it is possible to approximate to the actual eigenvalue with necessary accuracy. The convergence of the iterative process has been proved yet. A general criterion of partition into master and slave coordinate sets is also found. Some results of application of the method are reported to demon?. strate its effectiveness.