The daily tourist arrival time series has strong nonlinearity and instability which is hard to model. Therefore, the improvement of accurate daily tourist arrival forecasting is a problem worth studying. For this purpose, this research introduces a new combined forecasting model based on clustering. The original sequence is divided by fuzzy c-means (FCM) clustering into sub-sequences with similarities according to seasonality and day type of time series; then the sub-sequences are modeled and forecast using radial basis function (RBF) neural networks. Experimental results of historical daily tourist arrival data of Jiuzhaigou Valley and Gulangyu Island show that the forecast accuracy of the proposed FCM–RBF model achieved the best performance compared with other benchmark models.