Kupper et al. (1986, Biometrics 42, 85-98) considered the fitting of dose-response regressions to litter proportions in teratology experiments. They found that the estimators which maximise the beta-binomial likelihood become biased when the intralitter correlation is incorrectly assumed to be homogeneous. This paper considers the nature of this bias and its implications for the analysis of litter proportions in toxicology.