With the introduction of the CBOE Market Volatility Indices VIX, VXO, and VXN, practitioners have been given an alternative means to gage implied volatility and to forecast S&P 500, S&P 100, and NASDAQ 100 index return volatilities. We measure the forecasting performance of these volatility indices in single-factor and multi-factor regression models, as well as within an ARCH framework. Results are compared to other commonly applied historical standard deviation and conditional volatility based models. We find that implied volatility, measured in form of a volatility index, serves as a better predictor than past realized volatility. The best forecasting model is the single-factor volatility index model, followed by the Exponential Smoothing model, which worked particularly well during times of high stock market volatility.