This paper proposes a neural network ensemble model for classification of incomplete data. In the method, the incomplete dataset is divided into a group of complete sub datasets, which is then used as the training sets for the neural networks. Compared with other methods dealing with missing data in classification, the proposed method can utilize all the information provided by the data with missing values, maintain maximum consistency of incomplete data and avoid the dependency on distribution assumption. Experiments on two UCI datasets show the superiority of the algorithm to other two typical treatments of missing data in ensemble learning