Model integration is one of the core components for the realization of model-driven engineering. In particular, the seamless exchange of models among different modeling tools is of special importance. This exchange is achieved by the means of model transformations. However, the manual definition of model transformations is an error prone and cumbersome task. So matching techniques, originally intended for database schema integration, have been reused. The result is unsatisfactory as current matching approaches typically produce only one-toone alignments which are inappropriate for many integration problems. As a consequence, a detailed review and a manual post-processing step is often necessary. To tackle these problems, we propose the self-tuning framework SmartMatcher for improving automatically generated transformations. Our approach combines the power of an executable mapping language for bridging structural heterogeneities with the strength of an instance based quality evaluation model. In an iterative, feedback-driven process a mapping between two schemas is constructed and repeatedly enhanced.