C.B. Martin has shown that the simple conditional analysis of disposition concepts (x is disposed at time t to give response r to stimulus s) is in error. This is due to finkish dispositions which are caused to disappear by the stimulus s. David Lewis has proposed an improved analysis which takes account of finkish dispositions by requiring that the appropriate causal basis remains for a sufficiently long time. I argue that Lewis’ analysis also fails, because of the existence of antidotes. An antidote to a disposition interferes with its normal operation so that the stimulus does not bring about the usual response. I consider several possible defences of Lewis’ analysis and a plausible repair, but find these unsatisfactory. I conclude by suggesting that an analysis of disposition concepts is not available because an unavoidable indexical element (e.g., reference to normal circumstances) is present in explanations of these concepts. In this regard they may be thought of as akin to theoretical or natural kind concepts.