Integral membrane proteins often exhibit long hydrophobic stretches in their amino acid sequence that indicate membrane-spanning a-helices and therefore provide useful topological information. Accordingly, the algorithm of Kyte and Doolittle (1982), which determines the mean hydrophobicity within a sliding window, is routinely applied to new membrane protein sequences to identify transmembrane helices. There is, however, another class of membrane proteins, exemplified by bacterial porins, that are folded predominantly into @-pleated sheet structures and show an overall hydrophobicity similar to soluble proteins.The recent crystal structure elucidations of matrix porin and phosphoporin from Escherichia coli (Cowan et al., 1992) and the unrelated porin from R . capsulatus (Weiss & Schulz, 1992) prompted us to reevaluate existing prediction methods for transmembrane @-strands. All of these porins are folded into 16-stranded antiparallel @-barrels (Kinemage 1) that are associated to homotrimers. The parts of the 0-strands that are in contact with the hydrophobic core of the membrane are 7-9 residues in length with every second (i.e., external) residue being hydrophobic (Kinemage 2). The intervening, internal residues show no clear pattern and thus have no predictive value. These residues are found to be hydrophilic when being part of the pore lumen or hydrophobic when buried by the internal loop L3 (Cowan et al., 1992).The mean hydrophobicity of one side of a putative @-strand (0-side-hydrophobicity, H,) is obtained by averaging the hydrophobic indices h (Eisenberg et al., 1984) of every second residue within a sliding window (Vogel & Jahnig, 1986). A window width of 4 was used so that