Protein toxins are soluble molecules secreted by pathogenic bacteria which act at the plasma membrane or in the cytoplasm of target cells. They must therefore interact with a membrane at some point, either to modify its permeability properties or to reach the cytoplasm. As a consequence, toxins have the built-in capacity to adopt two generally incompatible states: water-soluble and transmembrane. Irrespective of their origin or function, the membrane interacting domain of most protein toxins seems to have adopted one out of two structural strategies to be able to undergo this metamorphosis. In the first group of toxins the membrane interacting domain has the structural characteristics of most known membrane proteins, I.e. it contains hydrophobic and amphipathic α-hellces long enough to span a membrane. To render this ‘membrane protein’ water-soluble during the initial part of its life the hydrophobic hellces are sheltered from the solvent by a barrel of amphipathic helices. In the second group of toxins the opposite strategy is adopted. The toxin is an intrinsically soluble protein and is composed mainly of β-structure. These toxins manage to become membrane proteins by oligomerizing in order to combine amphipathic β-sheet to generate sufflclent hydropho-bicity for membrane insertion to occur. Toxins from this latter group are thought to perforate the lipid bilayer as a β-barrel such as has been described for bacterial porins, and has recently been shown for staphylococcal α-toxin. The two groups of toxins will be described in detail through the presentation of examples. Particular attention will be given to the β-structure toxins, since four new structures have been solved over the past year: the staphyloccocal α-toxin channel, the anthrax protective antigen protoxin, the anthrax protective antigen-soluble heptamer and the CytB protoxin. Structural similarities with mammalian proteins implicated in the immune response and apoptosis will be discussed. Peptide toxins will not be covered in this review.