Finite impulse response (FIR) structures are the most commonly used digital filters and can be found in various areas of everyday life. In this paper we introduce a novel approach to construct and optimize a redundancy-free design of such filter structures. Relying on attribute grammars and derivation-tree based genetic programming the evolFIR system can restrict the search space to exactly those filter descriptions which fulfill each of the specified constraints. Furthermore, due to the sophisticated tree representation form, the resulting filter descriptions are not only valid, but also free of redundancy.