The author contends that in the state of nature we need some detectable property of informants that correlates well with their being right about p. This yields a twofold criticism of Robert Nozick's truth-tracking analysis of knowledge. First, it is not necessary that the informant be a good tracker in all close possible worlds, merely those that are open possibilities, those the inquirer cannot rule out as being non-actual. Second, the inquirer cannot set herself directly to pick out a good tracker of p, so Nozick's favoured knowledge-conferring property lacks the necessary epistemic accessibility. Keywords:epistemic accessibility,
Nozick,
possibility,
possible worlds,
truth-tracking analysis