We compare a large number of volatility models in terms of their ability to describe the behavior of the behavior of conditional variance, using out-of-sample data. Our question of interest is whether more sophisticated volatility models are able to outperform the simple GARCH(1,1) model. This question is addressed using the test for superior predictive ability (SPA) by Hansen (2001). Using DM-$ exchange rate data, we do not find evidence that the GARCH(1,1) is outperformed by other models. However, when we compare the models using IBM equity return data, we find the GARCH(1,1) to be significantly outperformed by alternative models. Most of the models that perform well in this data set are models that can accommodate a leverage effect. Our analysis confirms that the test for SPA of Hansen (2001) is more powerful that the Reality Check (RC) of White (2000).