In applications on mechanisms for metalloenzymes, the hybrid density functional method B3LYP has been used in most cases. The present knowledge about the accuracy of this method on transition metal complexes is described. In comparison to ab initio methods like CASPT2 and CCSD(T), B3LYP has generally been shown to perform quite well. However, there is one exception and this is for the type of copper dimer which is found in enzymes like hemocyanin, catechol oxidase and tyrosinase. Large deviations have been found between CASPT2 and B3LYP and also between B3LYP and experiments on model complexes. This situation is carefully investigated. The accuracy of B3LYP for the activation of O2 in enzymes is also analyzed by comparisons to experiments. For the barrier of O-O bond cleavage B3LYP appears to behave quite satisfactorily, while for the binding of O2 to the metal there are deviations compared to experiment. The question whether this is due to the B3LYP method or to the chemical models used is addressed.