What is the connection between pain and pain behaviour? Is it logically necessary, or is it contingent? Or is it too complex to be classified in terms of this Humean dichotomy? Surely it is too complex, for if we say the relationship is a necessary one, we should, apparently, have to deny that there could be pain without pain behaviour, or pain behaviour without pain; yet stoicism and shamming pain occur. If we say that the relationship is not necessary and so contingent, we should have to say that the natural correlate or expression of pain might have been languor or jolly laughter, and, conversely, that wincing, grimacing, writhing, groaning, etc., might have been the natural correlate or expression of sensations quite different from pains. To avoid the first difficulty it could be said that necessary connections need not be invariant. To avoid the second, it might be pointed out that some contingent connections are less contingent than others, and that, in particular, causal relationships need not always be of the external, Humean kind. But even if these observations are correct they are difficult precisely because they attempt to soften and blur a distinction whose hardness and exhaustiveness is constitutive of its clarity and strength. If we modify Hume's law we may do so only after a careful examination of a case which forces us to do so.