This study evaluates the accuracy of three competing volatility models : Historical Volatility, Exponentially Weighted Moving Average (EWMA), and Generalized Autoregressive Conditional Heteroscedastic (GARCH) in forecasting future daily volatility for SET50 INDEX over the period from 2 January 2008 to 30 December 2010. Volatility forecasts are evaluated by root mean squared error (RMSE) in comparison to 30-day, 90-day, and 180-day realized volatility. The finding suggests that GARCH (1,1) provides the most accurate forecast for 30-day realized volatility while EWMA is superior in forecasting 90-day and 180-day realized volatility. In addition, all models provide high forecasting error in the abnormal market condition. The implication is that both GARCH (1,1) and EWMA outperform Historical Volatility for 30-day and 90-day realized volatility but the performance of forecasting model is still varying for 180-day realized volatility. These results are useful in selecting the model for volatility forecasting in investment decision such as risk measurement and Option pricing.