Various techniques have been proposed to repair inconsistent relational data that violate functional dependencies by optimizing the repair plan by the metric of repair cost. However, they may fall short in the circumstances where the erroneous data occurs in the left-hand side of a functional dependency or repair cost is not a reliable optimization indicator. In this paper, a novel repairing approach based on possible world model is proposed. It first constructs candidate repair plans and then estimates their possible world probabilities. The possible world probabilities are measured by quantifying both repair cost and candidate value appropriateness with regard to other related attribute values presented in relational data. Finally, extensive experiments on synthetic datasets show that the proposed approach performs considerably better than the cost-based approach on repair quality.