The empirical pricing kernels estimated from index options are non-monotone (Rosenberg and Engle, 2002; Bakshi, Madan, and Panayotov, 2010) and the corresponding risk aversion functions can be negative (Ait-Sahalia and Lo, 2000; and Jackwerth, 2000). We show theoretically that these and several other properties of empirical pricing kernels are consistent with rank-dependent utility model with probability weighting function which over-weights tail events. We also estimate the pricing kernels non-parametrically from the S&P 500 index options and construct empirical probability weighting functions. The estimated probability weights typically have the inverse-S shape which overweights tail events and is widely supported by the experimental decision theory.