Nonribosomal peptide synthetases (NRPSs) catalyze the synthesis of a large group of structurally diverse natural products. Primary structure, size and complexity of a peptide product are dictated by the number and organization of iterated modules and domains, which constitute the protein template. Due to their modular organization NRPSs are predestinated for a rational protein design. Changing the protein template allows the generation of novel peptide derivatives, but in the past such manipulations often lead to a radical decrease of the productivity of the artificial systems, which is ascribed to a disturbed communication. This limitation might be overcome by protein evolution of NRPSs, although their enormous sizes raise other challenges. The basis for the construction of model systems for the evolution of NRPSs was established in this study. First, a system for in vitro evolution of the substrate selectivity of adenylation-(A)-domains was arranged. Following a similar study on t-RNA synthetases, an A-domain (from the TycA protein) was constructed in which almost all residues forming the substrate binding pocket were changed to alanine, leading to 15 single and multi-ple mutants. These were investigated in regard to their catalytic efficiency in activating the cognate amino acid L-phenylalanine as well as to their ability of activating mis- and non-cognate substrates. An observed site specificity for 3-nitro-tyrosine was used for the implementation of the in vitro compartmentalization-(IVC)-system. It was shown that all partial steps of the IVC-system were established successfully. Merely the final detection of the enzyme bound 3-nitro-tyrosine failed due to an insufficient specificity of the available antibodies. In the second part of this study, an in vivo method for the evolution of NRPSs was developed. It was based on the construction of different dimodular NRPS systems for the heterologous production of the cyclic dipeptide D-Phe-L-Pro-diketopiperazine (DKP). This dipeptide shows diverse bioactive properties e.g. as biosensor or antibiotic that can be used for the establishment of an in vivo selection system. The generated NRPS systems were determined in the host Escherichia coli and optimized with regard to their efficiency of product formation. By varying endo- and exogenous parameters, the productivity of the heterologous host E. coli for the nonribosomal synthesis of D-Phe-L-Pro-DKP could be increased about 400 % to 38 µM (12 mg/g BDW). This concentration of the dipeptide is sufficient for the generation of an in vivo selection system based on the antibiotic effect of DKP.