Ensemble learning has become a hot topic in machine learning. It dramatically improves the generalization performance of a classifier. In this paper, neural network ensemble for small data sets is studied and an approach to neural network ensemble (Novel\-NNE) is presented. For increasing ensemble diversity, a diverse data set is generated as part training set in order to create diverse neural network classifiers. Moreover, different combinational methods are studied for Novel\-NNE. Experimental results show that Novel\-NNE for both the relative majority vote method and the Bayes combinational method achieves higher predictive accuracy.