The effect of 12 salts on tobacco mosaic virus protein polymerization has been tested by means of sedimentation velocity and circular dichroism at pH 8 and various ionic strengths (0.01–0.80 M) and temperatures (15–30°C). The ‘phase diagram’ describing the polymorphism of TMV protein (Durham, A.C.H., Finch, J.T. and Klug, A. (1971) Nature, New Biol. 229, 37–42) is shown to be valid under the special conditions of its setup, both regarding the positions of phase boundaries between polymorphs, and the types of polymorph formed. In general, however, phase boundaries are determined by the position of the salt in the Hofmeister series. With increasing ionic strength and without CI −, s 20,w values rise to a first plateau near 8 S and a second plateau near 27 S; these aggregates form within minutes or faster. 20 S disks (suggested by the phase diagram) are not observed, at any time. With CI −, at 20–22°C, the 8 S plateau shifts to 16 S, particularly in the presence of Ca 2+ or Mg 2+; no 27 S species are observed. 16 S aggregates are not simulated by an interacting 8 S/20 S system; they do resemble, however, the A-protein oligomers (4 S/8 S). With NaCl or KCI (used for the phase diagram), a transient 8 S plateau is attained at intermediate ionic strengths, giving way to 20 S disks after several hours. With increasing temperature, the action of CI − (with or without Ca/Mg 2+) approaches that of the other salts, replacing 16 S by 8 S, and 20 S by 27 S; near 30°C, any 27 S formation proceeds via an intermediate state. Combining CD data with published X-ray data, the results are rationalized considering the hydrophobicities and flexibilities of the intersubunit binding sites. Temperature and ionic strength favour the lateral bonds; the axial bonds between two layers (‘pairing’) are weakened with increasing temperature, but they appear to be stronger in the 20–27 S aggregates, and in the presence of chlorides. ‘Cumulated strain’ (Kellenberger, E. (1972) Ciba Found. Symp. 7, 295–299), more effective in three-layer than in two-layer oligomers, is most probably the factor which limits the A-protein aggregation near 8 S and 16 S, respectively. This enables A-protein material of optimum size and flexibility (8 S) to be incorporated into ring structures (20–27 S), whereas 16 S material can be regarded as an aberrant two-layer form blocking ring formation. Possible implications for the virus assembly are discussed, by comparing the structural features of ‘disks’, at pH 7 and 8, and of A-protein.