This study evaluates forecasting accuracy among several competing methodologies, including time series and econometric methods, on visitation to 255 US National Park Service (NPS) sites. The performance of these models is contrasted with the model currently in use by the NPS. One-yearahead, 2-year-ahead, and combined (1-and 2-year-ahead) forecasting performance at the individual park level is examined utilizing several measures of forecasting accuracy, including root mean square error (RMSE) and mean absolute percentage error (MAPE). Results indicate incorporating economic variables can significantly improve forecasts, particularly for large and small parks. For medium size parks the naive forecast errors were typically lowest. Furthermore, the naive model performed well, often producing the best forecast, followed by the econometric model. Regionally, the naive and econometric models preform best, with the Pacificwest region being the exception. Utilizing the most accurate model for each park leads to a 24% improvement over current forecasts (1-year horizon) and suggests that a mixed model approach is optimal. Pergrams and Zaradic (2006) demonstrate that per capita use of US national parks has declined since the late 1980s. Additionally, The Economist (2008) has highlighted the fact that Americans are retreating from such outdoor activities. The significant economic impacts of, and potential downward trends in, national park visitation combine to make the accuracy of visitation forecasts a principal concern. The US NPS currently provides forecast estimates for approximately 350 parks, and understanding future demand is critical to the provision of services, appropriate application of funds, and budgeting concerns.