Background: Pathways provide important information about how genes interact with each other in a concerted way.Nevertheless, our knowledge about them is fragmented.Although biological experiments offer a reliable way for pathway expansion, they are expensive and time consuming.Several previously developed computational methods for pathway expansion focused on analysis and modeling of individual experimental datasets, and they seldom fully utilized the rapidly accumulated prior knowledge in their inferences.Hence, their pathway expansion results were often not satisfactory and also lacked sounding biological interpretations.Methods: In this study, we developed a knowledge-mining based algorithm for pathway expansion with protein-protein interaction data and gene ontology.First, we used proteinprotein interaction data documented in HPRD database to identify the interacting neighbor genes for a target gene, and used GO (Gene Ontology) structure to define the distances for measuring functional similarities between the target gene and its neighbors.Then, we found the nearest neighbor genes for the target gene.Since two genes that are very similar in GO functions are very likely to take part in the same pathways, we predicted the target genes pathways as its nearest neighbor genes.Results: Totally, we analyzed 3937 genes.On average, 64.51% of the known pathways were correctly predicted by our algorithm.Furthermore, we also evaluated the capability of our algorithm to predict novel pathways.Of seven genes whose pathways were unknown in March 15, 2011, four genes turned out to be consistent with our predictions, verified by using the updated knowledge released on March 18, 2012.Conclusions: This study shows that the proposed knowledge-mining based algorithm offers an alternative and improved avenue to expand the current pathways .