<p>I explore a problem for the standard process reliabilist analysis of proper inference in terms of conditional reliability. The difficulty involves Goodman’s new riddle of induction, which can be used to argue that inductive inferences are in fact conditionally unreliable, and hence that no inductive inference is capable of producing a justified belief given the conditional reliability analysis. Accepting this conclusion would commit reliabilism to an unattractive form of inductive skepticism. I then consider some ways a reliabilist could resolve this problem, finally proposing a new analysis incorporating inferential dispositions.</p>