I propose a novel method to reliably estimate the conditional pricing kernel by incorporating conditioning variables. The VIX and the term spread are conditioning variables that are most informative about state prices. The conditional kernel estimate exhibits significant time variation: the more favorable market expectations, the higher state prices in negative return states. During bad times, the equity premium implied by the conditional kernel is entirely attributable to compensation for left-tail scenarios, contrasting with findings from the unconditional kernel. The out-of-sample return forecast based on the conditional kernel has a higher R-square than the one based on the unconditional kernel.