Recently, a new method for the specific covalent labeling of fusion proteins in vitro and in living cells has been developed. This method is based on the unusual mechanism of a DNA repair protein: the O6-alkylguanine DNA alkyltransferase (AGT). In addition to its natural substrate, O6-alkylated guanine incorporated in DNA, the AGT protein can transfer the benzyl group of O6-benzylguanine to a reactive cysteine residue, leading to its irreversible alkylation. The labeling system relies on the fact that affinity or fluorescent tags can be attached to the para position of the benzyl ring and thus be transferred to the AGT protein, leading to its covalent labeling. Although this system has already been successfully applied, it still suffers from two major drawbacks: the low reactivity of the protein toward O6-benzylguanine derivatives and the limitation to cells deficient in endogenous AGT. The work reported here describes directed evolution strategies that were developed to overcome these drawbacks. The first goal was to increase the labeling efficiency in living cells by increasing the reactivity of AGT toward O6-benzylguanine derivatives. Structural data available on AGT and the results of docking experiments pointed out four residues, 140, 157, 159 and 160, which are close enough to the O6-benzylguanine to allow interactions with this substrate. Therefore, these four positions were fully randomized and selections were performed using a phage display system based on the phagemid technology. Two mutants, PGEAhAGT and PGEGhAGT, showed a 15- and 20-fold increase in the reaction rate in vitro. It has subsequently been shown that this increase in reaction rate in vitro is also reflected in living cells by a more efficient labeling of AGT fusion proteins. In conclusion, the selected mutant proteins allow for a highly efficient covalent labeling of AGT fusion proteins in vitro as well as in living cells and therefore should become an important tool for studying protein function. The second aim was to enable the selective labeling of AGT fusion proteins in living cells in the presence of endogenous AGT. For this purpose, four positions, 131, 132, 134 and 135, which are believed to interfere with N9-substituted O6-benzylguanines were simultaneously completely randomized. Phage display selections allowed the identification of a clone, AGT54, with increased reactivity toward O6-benzylguanine and complete resistance toward an N9-substituted O6-benzylguanine. This mutant was further engineered to render the protein more stable, non-DNA binding and to decrease the overall protein size, leading to the so-called MAGT. This new mutant therefore allowed the specific labeling of fusion proteins in living cells in presence of endogenous AGT by simply pre-incubation with the N9-substituted O6-benzylguanine. In conclusion, the possibility to specifically label AGT fusion proteins in the presence of endogenous protein should significantly broaden the scope of application of AGT fusion proteins for studying protein function in living cells. Finally, a transcription-based yeast three-hybrid system was adapted and successfully employed to identify clones with a 2-fold increased activity toward O6-benzylguanine out of an error prone library based on MAGT.