Species of the genus Fusarium are well-known plant pathogens and mycotoxigenic fusaria are associated with health hazards to humans and animals. There is a need to understand the mechanisms of mycotoxin production by Fusarium species and to predict which produce mycotoxins. In this study, the Fusarium phylogenetic tree was first inferred among trichothecene producers and related species. We reconstructed the maximum likelihood (ML) tree based on the combined data from nucleotide sequences of rDNA cluster regions, the β-tubulin gene (β-tub) and the elongation factor 1α gene (EF-1α). Second, based on this tree topology, the ancestral states of the producing potential of type A and B trichothecenes (TriA and TriB), zearalenone (ZEN), moniliformin (MON), beauvericin (BEA) and enniatins (ENN) were reconstructed using the maximum parsimony (MP) method based on the observed production by extant species as reported in the literature. Finally, the species having the potential to produce each of these six mycotoxins was predicted on the basis of the parsimonious analysis. The ML tree indicated that the Fusarium species analysed in this study could be divided into two major clades. Clade I was divided into four distinct subclades: I-a, I-b, I-c and I-d. Furthermore, the parsimony reconstruction suggested that the potential for producing MON and ZEN was gained or lost only once, and that the producing potential for TriA and TriB, BEA and ENN was repeatedly gained and lost during the evolutionary history of the Fusarium species analysed in this study. Interestingly, the results showed the possibility that several species, about which reports were scarce with regard to mycotoxin production, have the potential to produce one or more of the six evaluated in this study. The phylogenetic information therefore helps one to predict the mycotoxin-producing potential by Fusarium species, and these "phylotoxigenic relationships" may be useful for predicting the pathogenicity of fungi.