Epistemic modals are expressions that quantify over possibilities about how the actual world is that are compatible with someone's information about it. There is a long-lasting disagreement among philosophers about the semantics of these expressions. For instance, the sentence 'it might rain' is taken to say that, as far as someone knows (relative to someone's knowledge), it is possible that it will rain. The disagreement then surrounds the question of whose knowledge is relevant in this instance (e.g. the speaker's knowledge, the assessor's knowledge, knowledge of the participants in the conversation). Nevertheless, in the background of this disagreement, there is a tacit agreement that epistemic modals are expressions referring to epistemic possibilities relativised to someone's knowledge. In my talk, I will throw doubt on this view and defend instead the understanding-relative account of epistemic modals. Firstly, I will argue that, because of being an inferentially structured and non-factive attitude that comes in degrees, understanding can help us in resolving such issues as the translatability of evidential epistemic modals in the languages other than English, unreliable claims about epistemic possibilities or the problem of ignorance (Dietz, 2008). Upon that, I will attempt to show that the understanding-relative analysis deserves more attention also because it re-frames the problems faced by the two dominant semantic approaches to epistemic modals (standard contextualism and relativism). On the one hand, I will conclude that the analysis creates problems for group contextualism (a type of standard contextualism that is immune to the objection from rejection and retraction) because of the inscrutability of collective understanding as a sensible epistemic attitude. On the other hand, I will argue that the analysis seems to be useful in dealing with the problems related to relativism.