This paper investigates the combination of individual forecasting models and their roles in improving forecasting accuracy and proposes two non‐linear combination forecasting models using Radial Basis Function and Support Vector Regression neural networks. These two non‐linear combination models plus the standard Multi‐layer Perceptron neural network‐based non‐linear combination model are examined and compared with the linear combination models. The UK inbound tourism quarterly arrival data is used and the empirical results demonstrate that the proposed non‐linear combination models are robust and outperform the linear combination models that currently dominate in the tourism forecasting literature. Copyright © 2013 John Wiley & Sons, Ltd.