This paper discussed the information content of threat signals. It was an attempt to explain the inconsistencies between the points of view of model-builders using game theory as a tool (represented in a paper by CARYL in 1979 in this journal), and of the (mainly field-) ethologists spending a considerable part of their time observing animals in groups (expressed for instance in the paper by BOSSEMA & BURGLER in this issue). Arguments were presented for the transfer by threat signals of both information about intentions (motivation), and information about "resource-holding potential" (strength, ownership, etc.). Individual recognition was expected to be associated with honest signals about intentions. CARYL'S deviating findings could not be attributed to an absence of individual recognition in the animals he considered. His findings could also not be explained very well by the fact that he hardly considered subtle signals, although the present paper argued that information about intentions is mainly given by subtle signals, and information about resource-holding potential by elaborate action patterns. Imperfect methods in the papers cited by CARYL were considered as the most important source for the deviations. Finally it has been discussed to what extent observational data as presented by BOSSEMA and BURGLER help in solving problems raised by the model-builders. The occurrence of frequent, short escalations has been suggested as a mechanism for preventing bluff. The evolution of graded warning-signals could be related to (1) the low cost of a warning as compared to an attack, and (2) the settlement of a conflict on the basis of differing motivations. This paper considered the possibility of heterogeneous summation of (already known) information about the resource-holding potential of an opponent, and the information about its intentions (from the displays).