Background The ability to accurately predict enzymatic functions is an essential prerequisite for the interpretation of cellular functions, and the reconstruction and analysis of metabolic models. Several biological databases exist that provide such information. However, in many cases these databases provide partly different and inconsistent genome annotations. Description We analysed nine prokaryotic genomes and found about 70% inconsistencies in the enzyme predictions of the main annotation resources. Therefore, we implemented the annotation pipeline EnzymeDetector. This tool automatically compares and evaluates the assigned enzyme functions from the main annotation databases and supplements them with its own function prediction. This is based on a sequence similarity analysis, on manually created organism-specific enzyme information from BRENDA (Braunschweig Enzyme Database), and on sequence pattern searches. Conclusions EnzymeDetector provides a fast and comprehensive overview of the available enzyme function annotations for a genome of interest. The web interface allows the user to work with customisable weighting schemes and cut-offs for the different prediction methods. These customised quality criteria can easily be applied, and the resulting annotation can be downloaded. The summarised view of all used annotation sources provides up-to-date information. Annotation errors that occur in only one of the databases can be recognised (because of their low relevance score). The results are stored in a database and can be accessed at http://enzymedetector.tu-bs.de.