This paper is not directly in the service of either of the ambitious analytic projects of analysing causation in terms of the holding of certain counterfactuals, nor of analysing counterfactuals in terms of causal matters. It focuses instead on one of the traditional puzzles that connect the two that arise almost whatever one takes the connection between counterfactuals and causation to be. The puzzle has no neat label that I am aware of, but it arises in its clearest form when we consider counterfactuals involving antecedent states that involve a difference from the actual course of events at a particular time, and a consequent, at least in part involving a state somewhat later than the time of the antecedent difference. In the possible worlds framework, the puzzle is often put in terms of what other differences there in the relevant possible worlds where the antecedent is true. Do those worlds match ours with respect to nearly all their pasts until the time relevant to the antecedent? Do they require 'small miracles' relative to the laws of the actual world? Is their causal structure the same except for an 'intervention' on a state associated with the antecedent? . . . and so on. Let me label this problem the 'deviation problem' to suggest that it concerns what deviations from actuality would be required for the antecedent to be true, in the class of counterfactuals of interest.In this paper I will propose a novel solution to the deviation problem. This solution will have several signal advantages over a number of the better-known proposed solutions to this problem, though it also incurs some distinctive costs. I am not sure, then, whether something like it will turn out to be the best solution, though I am sure it deserves a run for its money alongside its better-