The Knobe effect (Analysis 63(3):190-194, 2003a) consists in our tendency to attribute intentionality to bringing about a side effect when it is morally bad but not when it is morally good. Beebe and Buckwalter (Mind Lang 25:474-498, 2010) have demonstrated that there is an epistemic side-effect effect (ESEE): people are more inclined to attribute knowledge when the side effect is bad in Knobe-type cases. ESEE is quite robust. In this paper, I present a new explanation of ESEE. I argue that when people attribute knowledge in morally negative cases, they express a consequence-knowledge claim (knowledge that a possible consequence of an action is that harm will occur) rather than a predictive claim (knowledge that harm will actually occur). I use the omissions account (Paprzycka in Mind Lang 30(5):550-571, 2015) to explain why the consequence-knowledge claim is particularly salient in morally negative cases. Unlike the doxastic heuristic account (Alfano et al. in Monist 95(2):264-289, 2012), the omissions account can explain the persistence of ESEE in the so-called slightchance of harm conditions. I present the results of empirical studies that test the predictions of the account. I show that ESEE occurs in Butler-type scenarios. Some I would like to thank two anonymous reviewers for a very careful reading of the text and for making valuable criticisms and suggestions that made rethink many of the issues discussed in the paper. I am very grateful to Bartosz Maćkiewicz who not only helped to implement the studies in Lime Survey, ran the studies, carried out or served as a consultant on many statistical analyses, but who also put forward constructive criticisms. Many thanks are due to Katarzyna Kuś (for all kinds of help), Michał Barcz,