This paper presents novel data on the German discourse particle 'wohl', which has been analyzed as a marker of uncertainty by Zimmermann (2008), and argues for treating 'wohl' as an inferential evidential. The argument is twofold. First, in declaratives 'wohl' is felicitous in contexts the respective modified proposition is known to be true, which is incompatible with an account in terms of uncertainty. Second, the distribution of 'wohl' in interrogatives is more complex and more restricted than assumed by the standard account: Following Truckenbrodt (2006), I assume that V2-interrogatives are undirected questions that can be licensed by 'wohl' but whose undirectedness effect is independent of 'wohl'. V-final interrogatives, on the other hand, are canonical directed questions but can only host 'wohl' when targeting content that cannot be known directly. The final analysis is couched in the framework of Murray (2010), proposing that 'wohl' contributes a not at-issue restriction of the common ground to those worlds in which the speaker (in declaratives) or addressee (in interrogatives) has inferential evidence.