This paper investigates the option implied risk measures–volatility, skewness and kurtosis–by applying the principle of maximum entropy. This approach does not rely on any parametric model, and allows to measure higher moments such as skewness and kurtosis, whereas traditional Black-Scholes model assumes zero skewness and constant kurtosis. Different from other model-free methods, e.g. Bakshi and Madan (2003), our approach does not require a large number of options with strike prices covering the entire support of the return distribution. Furthermore, this method provides confidence interval for option implied risk measures. Simulations show that the entropy approach outperforms the Black-Scholes model and the model-free method in backing out the implied volatility, particularly when the risk neutral distribution possesses heavy tails and non-zero skewness. Using S&P500 index options, we apply our method to obtain implied volatilities and test its forecasting performance. We show that the implied volatility obtained from our method subsumes all information in the Black-Schole implied volatility and historical volatility. In addition, it has more predictive power than the model-free implied volatility following Bakshi and Madan (2003), in both in-sample and out-of-sample setup.