A number of model transformation approaches have been proposed both from academia and industry since automated manipulation of models plays a central role in model driven development. Ideally, a model transformation technique should also be compatible with manual changes that might be performed by designers on the generated models in order to resolve unforeseen requirements or limited expressiveness of the involved metamodels.

This paper proposes an approach to model transformation based on answer set programming. Starting from target models that have been manually modified (and possibly not belong to the co-domain of the transformation being used), the approach is able to deduce a collection of models that approximate the ideal one from which it is possible to generate the previously modified target.