We investigate interchangeability of values in CSPs, based on an approach where a single value in the domain of a variable can be treated as a combination of “sub-values”. An algorithm for removing overlapping sub-values is presented. The resulting CSPs take less time to find all solutions and yield a more compactly-representable, but equivalent, solution set. Experimental results show that, especially in loose problems with large numbers of solutions, dramatic savings in search cost are achieved.