The bootstrap, the jackknife, and classical methods are compared through their confidence intervals for the proportion of affected fetuses in a common type of animal experiment. Specifically, suppose that for the ith of M pregnant animals, there are x i affected fetuses out of n i total in the litter. The conditional distribution of x i given n i is sometimes modeled as binomial (n i p i ), where p i is a realization from some unknown continuous density. The p i are not observable and it is of interest in some toxicological experiments to find confidence intervals for E(p). Theory suggests that the proposed parametric bootstrap should produce higher order agreement between the nominal and actual coverage than that exhibited by the usual nonparametric bootstrap. Some simulation results provide additional evidence of this superiority of the modified parametric bootstrap over the jack-knife and classical approaches. The proposed resampling is flexible enough to handle a more general model allowing correlation between p i and n i .