MOTIVATION: Genome-scale metabolic network reconstructions have been established as a powerful tool for the prediction of cellular phenotypes and metabolic capabilities of organisms. In recent years, the number of network reconstructions has been constantly increasing, mostly because of the availability of novel (semi-)automated procedures, which enabled the reconstruction of metabolic models based on individual genomes and their annotation. The resulting models are widely used in numerous applications. However, the accuracy and predictive power of network reconstructions are commonly limited by inherent inconsistencies and gaps. RESULTS: Here we present a novel method to validate metabolic network reconstructions based on the concept of autocatalytic sets. Autocatalytic sets correspond to collections of metabolites that, besides enzymes and a growth medium, are required to produce all biomass components in a metabolic model. These autocatalytic sets are well-conserved across all domains of life, and their identification in specific genome-scale reconstructions allows us to draw conclusions about potential inconsistencies in these models. The method is capable of detecting inconsistencies, which are neglected by other gap-finding methods. We tested our method on the Model SEED, which is the largest repository for automatically generated genome-scale network reconstructions. In this way, we were able to identify a significant number of missing pathways in several of these reconstructions. Hence, the method we report represents a powerful tool to identify inconsistencies in large-scale metabolic networks. AVAILABILITY AND IMPLEMENTATION: The method is available as source code on http://users.minet.uni-jena.de/ approximately m3kach/ASBIG/ASBIG.zip. CONTACT: email@example.com SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.