In this study we compare a set of Markov Regime-Switching GARCH models in terms of their ability to forecast the Tehran stock market volatility at different time intervals. SW-GARCH models have been used to avoid the excessive persistence that usually found in GARCH models. In SW-GARCH models all parameters are allowed to switch between a low or high volatility regimes. Both Gaussian and fat-tailed conditional distributions are assumed for the residuals, and the degrees of freedom can also be state-dependent to capture possible time-varying kurtosis. Using stationary bootstrap and re-sampling, the forecasting performances of the competing models are evaluated by statistical loss functions. The empirical analysis demonstrates that SW-GARCH models outperform all standard GARCH models in forecasting volatility. Also, the SW-GARCH model with the t distribution for errors has the best performance in fitting a model and estimation.