The handicap theory, in which the cost of waste guarantees honest adve
rtising, is being used increasingly in solutions to the problems of bi
ological signal evolution. However, it is usually applied to systems w
hich are insufficiently understood to allow testing against alternativ
e theories. In particular, the ability of the handicap theory to expla
in the design of signals has never been properly tested. We test its a
bility to explain signal design features in an unusually well studied
area of biological signalling: warning coloration and mimicry. Since a
full handicap model proves immediately unrealistic, we modify the mod
el to incorporate realistic assumptions about predator learning. Using
this model we explicitly compare the handicap theory with a purely ''
conventional'' signalling model and with a null model. Predictions rel
ating to three key design features (conspicuousness, pattern similarit
y, and Batesian mimicry) are compared, and tested against available da
ta. Although many predictions remain to be tested adequately, we concl
ude that: (i) conspicuousness is most plausibly explained by the conve
ntional signalling theory that ascribes the function of conspicuous co
loration to signal efficacy rather than waste; (ii) pattern similarity
, within and between species, is unlikely to be the result of the need
to produce similar degrees of conspicuousness, as predicted by the ha
ndicap theory, but is plausibly explained as the result of pattern gen
eralization amongst discriminating predators, as predicted by the conv
entional signalling theory; and (iii) Batesian mimicry is predicted by
the conventional signalling theory, but not the handicap theory. Ther
efore the handicap theory fails to provide an adequate explanation of
the main design features of at least one major signalling system.