In this study, four steps from Box-Jenkins were used to set up ARMA model and ARIMA model, and the method of least squares regression was used to analyze and predict the number of visitors to national scenic areas, as well as show the performance of a combination of forecasting models. According to regression analysis, this study showed that CPI, GDP, seasonal indices, the number of tourists in the previous year, and the number of tourists in the same period in the previous year have a significant impact on the number of tourists to national scenic areas. Increase in income helped to increase people's willingness to travel, and price index helped to increase spending power during tourism seasons. The result of this study also indicated that the third quarter is relatively a peak season for Taiwan's national scenic areas and the dissemination of information had a positive impact on the number of tourists. In terms of forecast performance, results from the MAPE values, Taylor coefficients, R2 and adj R2, showed that accuracy can indeed improve through a combination of forecasting models.