We describe a new approach to find models for a computational higher-order logic with datatypes. The goal is to find counterexamples for conjectures stated in proof assistants. The technique builds on narrowing [14] but relies on a tight integration with a SAT solver to analyze conflicts precisely, eliminate sets of choices that lead to failures, and sometimes prove unsatisfiability. The architecture is reminiscent of that of an SMT solver. We present the rules of the calculus, an implementation, and some promising experimental results.