Abstract: Feral horse, Equus caballus, breeding groups, called bands, usually include one but sometimes up to five stallions. We found that mares were loyal to single-stallion (SS) or multistallion (MS) bands or were social dispersers (maverick mares, Mv). The spacing and social behaviour of mares and stallions in single- and multistallion bands was measured. Indices of mare well-being were also measured including activity budgets (feeding: MS>SS=Mv; resting: MS<SS=Mv), band and mare travel (MS>SS), maternal effort in maintaining contact with foals (MS=Mv>SS), parasite levels in faeces (MS>Mv>SS), body condition (MS=Mv<SS), fecundity (Mv<MS<SS) and offspring mortality (Mv<MS<SS). We present evidence suggesting that the poorer well-being of maverick mares and multistallion band mares results from greater harassment by stallions. Stallion and mare behaviour and poor reproductive success in multistallion bands were not consistent with explanations for the existence of such bands based on cooperation or alternative mating strategies. We suggest an alternative explanation. Stable relationships between mares and a single stallion may enhance reproductive success by reducing aggression between individuals. Therefore, we propose that there is strong selection pressure for stable, long-term stallion-mare relationships, called consort relationships. We propose the consort hypothesis, that multistallion bands are an artefact of selection for stable relationships that occasionally result in more than one such relationship forming, because mares solicit more than one stallion and stallion dominance changes during band formation. Copyright 1999 The Association for the Study of Animal Behaviour.