Extreme learning machine(ELM)is an algorithm for training single-hidden layer feed-forward neural-network (SLFN)with fast speed and good generalization. It is indispensable to firstly select an appropriate architecture of SLFN when applying ELM to solve practical problems. However. It is very difficult to select an appropriate architecture of SLFN. In order to deal with this problem,an ensemble learning method is proposed in this paper. It is unnecessary to select the appropriate architecture of SLFN when using the proposed method to solve the problems. The proposed method includes 3 steps:(1)initialize a big SLFN;(2)retrain some SLFNs with dropout hidden nodes with ELM;(3)the trained SLFNs are integrated by majority voting method,and the integrated model is used to classify testing instances. We experimentally compared the proposed approach with traditional ELM on 10 data sets,and the experimental results confirm that the proposed apprach outperforms the traditional ELM on performance of classification.