Zahavi's handicap theory, formalized by Grafen, suggests that 'cheater
s' must be at a disadvantage if a communication system such as rituali
zed aggression is to evolve (Grafen 1991, In. Behavioural Ecology: An
Evolutionary Approach (Ed. by J. R. Krebs & N. B. Davies), pp. 5-31. O
xford: Blackwell Scientific). To determine whether cheating is disadva
ntageous in Betta splendens, we held a series of live interactions, af
ter inducing hyper-aggression by socially isolating and then briefly '
priming' the fish. Primed isolates, which were no stronger than their
rivals, 'cheated' by escalating rapidly to tailbeating and biting. The
se cheaters, however, usually lost fights to non-isolated opponents. U
nprimed isolates, i.e. socially isolated fish that were not primed, we
re not initially hyper-aggressive and thus did not cheat. They lost fe
wer fights than the cheaters. Results suggested that cheaters lost bec
ause they exhausted themselves by their hyper-aggressiveness, allowing
their non-hyper-aggressive opponents to win. This result is consisten
t with the Zahavi-Grafen model of how an 'honest' level of ritualized
aggression can be stabilized in a population. (C) 1998 The Association
for the Study of Animal Behaviour.