i. In this paper I examine the way in which predictions from a model based on games theory can help us understand ethological data about agonistic displays. 2. The War of Attrition is a model which explains how contests might be settled entirely by display. The model can be extended to deal with contests involving attack as well as display, assuming that the cost of attack is greater than the cost of display. It predicts that information about the aggressiveness should not be transferred in the display. 3. This prediction contrasts with the traditional ethological view that aggressive displays evolved to transmit information about the probability of attack or escape. 4. Data from the ethological literature on Parus species, Pheucticus ludovicianus and Stercorarius skua are examined in which both the sender's action after displaying, and the receiver's response, were recorded. 5. Attack could be predicted much less well than escape in all species. In contrast to the displays indicating escape, aggressive displays were not consistent in the probability of attack they signalled at different times, and were sometimes as likely to be followed by escape as by attack. It is concluded that information about attack is poorly encoded in the aggressive displays. 6. In three of four species, the most highly aggressive display had no effect on the behaviour of the receiver. Other displays did modify its behaviour. To find whether information about attack was responsible, the effects of this information must be separated from those of information about escape which is correlated with it in the displays. When this is done, the evidence suggests that information about attack is unimportant in affecting the behaviour of the receiver, while information about escape is important (at least in the Blue Tit and Great Skua). 7. The data support the prediction from the War of Attrition, and suggest that the importance of transferring information about escape should be examined theoretically. 8. The War of Attrition model also predicts the distribution of contest lengths. Contests between female Iguana agreed roughly with the prediction, but contests between female Betta splendens or male Molothrus ater did not. The distributions in these species can be understood if the cost of display is not linearly related to the length of the contest, and a generalized model is described. More work is required on the form of cost-functions. 9. In the discussion, the implications of the results for the methodology of ethological investigations of displays are reviewed, and alternative models from ethology and games theory briefly discussed. 10. Theoretical models are of necessity highly simplified, but may sometimes lead to predictions which throw new light on data. Ethologists should treat models as null hypotheses against which to compare their data, and should use the development of a model as an occasion to re-examine their data and assumptions.