Clathrin-coated vesicles were uncoated with the 70-kD 'uncoating ATPase' from bovine brain, and the molecular products were visualized by freeze-etch electron microscopy. This yielded images of released clathrin triskelia with up to three 70-kD uncoating ATPase molecules bound to their vertices. Likewise, incubation of soluble clathrin triskelia with purified uncoating ATPase also led to trimeric binding of the ATPase to the vertices of clathrin triskelia. However, this occurred only when either EDTA or nonhydrolyzable analogues of ATP were present, in which case the ATPase also appeared to self-associate. When ATP was present instead, no 70-kD ATPases could be found on clathrin triskelia and all ATPases remained monomeric. These observations support the notion that ATP controls an allosteric conversion of the 70-kD uncoating ATPase between two different molecular conformations, an ATP-charged state in which the molecule has relatively low affinity for itself as well as low affinity for clathrin, and an ATP-discharged state in which both of these affinities are high. We presume that in vivo, the latter condition is brought about by ATP hydrolysis and product release, at which point the ATPase will bind tightly to clathrin and/or self-associate. We further propose that these reactions, when occurring in concert within a clathrin lattice, will tend to destabilize it by a mechanism we call 'protein polymer competition'. We stress the analogies between such a mechanism of uncoating and the ATP-driven events in muscle contraction. Finally, we show that under experimental conditions in which the uncoating ATPase fully removes the coats from brain coated vesicles, identical aliquots of the enzyme do not affect plasmalemmal coated pits in situ. This remarkable selectivity, the mechanism of which remains a complete mystery, is at least consistent with the idea that the 70-kD ATPase indeed plays a role in uncoating coated vesicles after they have formed in vivo.