Recently, Yalcin (Epistemic modals. Mind, 116, 983–1026, 2007) put forward a novel account of epistemic modals. It is based on the observation that sentences of the form ‘({phi}) & Might ({ egphi}) ’ do not embed under ‘suppose’ and ‘if’. Yalcin concludes that such sentences must be contradictory and develops a notion of informational consequence which validates this idea. I will show that informational consequence is inadequate as an account of the logic of epistemic modals: it cannot deal with reasoning from uncertain premises. Finally, I offer an alternative way of explaining the relevant linguistic data.