Concrete semantics used for abstract interpretation analyses are generally expressed as fixpoints. Checking a property on this kind of semantics can be done by intersecting the fixpoint with a specification related to the property. In this paper, we show how to produce a new, “reverse” analysis from this specification. The result of this analysis, expressed as a lower closure operator, is then used to guide the initial analysis. With this approach, we can refine the result given by the direct abstract analysis.We show that this method enables to deduce forward analyses from backward analyses (and vice-versa), and to combine them iteratively in a way similar to the forward-backward combination of analyses.