Abstract: Accurate prediction of the tourism demand is important for tourism management. The state-of-art methods for predictive modeling used in tourism research include traditional statistical methods, soft computing methods, and artificial intelligence methods. Note that artificial intelligence methods, which were introduced to tourism research in the 1990s, have greatly improved the predictive accuracy of modeling methods. Machine learning is an important area of artificial intelligence, which has been widely recognized as a powerful tool for intelligent data analysis. BP neural network, as one of the traditional machine learning technique, has been widely used to construct predictive model for intelligent data analysis. However, BP neural network suffers from several drawbacks, such as overfitting, difficulties in setting parameters, and local minima problem, and hence the performance of BP neural network is very unstable in practical applications. This paper combines an advanced machine learning paradigm named ensemble learning with BP neural network to build neural network ensemble for tourism demand prediction. This study conducts predictive modeling for tourism demand of three important tourist source countries of US, Britain and Australia for travel to Chinese mainland. The results show that, by combining a number of diverse neural networks, neural network ensemble significantly improves the predictive accuracy over traditional statistical methods and traditional machine learning methods including single BP neural network. Such method provides a better choice for more accurate predictive modeling for tourism demand.