The clustered organization of Hox genes provides a powerful opportunity to examine gene gain and loss in evolution because physical linkage is a key diagnostic feature which allows homology to be established unambiguously. Furthermore, Hox genes play a key role in determination of axial and appendicular skeletal morphology and may be a key component of the evolution of diverse metazoan body forms. Despite suggestions that changes in Hox gene number played a role in evolution of metazoan body plans, there has been a general lack of evidence for such variation amongst gnathostomes (or indeed any vertebrate) and it has therefore been widely assumed that differential regulation may be the key element in all vertebrate Hox evolution. We have studied the Hox gene clusters of a teleost fish, Fugu rubripes, to test the possibility that Hox organization may have varied since the origin of jawed vertebrates. We have identified four Hox complexes in Fugu and found an unprecedented degree of variation when compared with tetrapod clusters. Our data show that: Fugu clusters are widely variant with respect to length; at least nine genes have been lost; there is a new group-2 paralogue; and pseudo-gene remnants of group-1 and group-3 paralogues were found in the Hoxc complex, when compared with the present mammalian clusters. We show that gene loss after duplication of the prototypical vertebrate Hox clusters is a key feature of both tetrapod and fish evolution.