The analysis of a variety of data sets (transcriptome arrays, phylogenetic profiles, etc.) yields groups of functionally related genes. In order to determine their biological function, associated gene groups are often projected onto known pathways or tested for enrichment of known functions. However, these approaches are not flexible enough to deal with variations or novel pathways. During the last decade, we developed and refined an approach that predicts metabolic pathways from a global metabolic network encompassing all known reactions and their substrates/products, by extracting a subgraph connecting at best a set of seed nodes (compounds, reactions, enzymes or enzyme-coding genes). In this review, we summarize this work, while discussing the problems and pitfalls but also the advantages and applications of network-based metabolic pathway prediction.