The Dutch scrapie eradication programme aims at the exclusive use of homozygous ARR/ARR breeding rams by the end of 2004. As a consequence, breeds with a small population size and a low frequency of the ARR allele may suffer unacceptable losses of genetic diversity and high inbreeding levels. We simulated three breeding strategies to assess their efficacy in eliminating scrapie sensitive alleles and their effect on inbreeding levels. Under mild selection, both homozygous and heterozygous ARR rams were used indiscriminately. Under moderate selection, homozygous ARR rams were used preferably but they were supplemented with heterozygous rams when necessary. Under severe selection, only homozygous rams were used. Severe selection mimics then the proposed eradication programme. Simulations were carried out with allelic frequencies and population parameters of existent rare breeds in The Netherlands. With severe selection all simulated breeds showed unacceptably high inbreeding rates (> 0·5% per year). For some breeds, moderate selection resulted in acceptable inbreeding rates, while for other breeds only mild selection resulted in acceptable rates. The frequency of the ARR allele after 5 years of selection was only slightly lower with moderate selection than with severe selection (0·8% lower on average), but it was clearly lower with mild selection. Based on these simulations, we propose a selection programme where with low frequencies of the ARR allele, mild selection is used initially. Once the ARR frequencies rise to a particular value, the switch can be made to moderate selection. The population size (e.g. below 750, 750 to 3750 and above 3750 ewes) determines the frequency of the ARR allele at which the switch can be made (33%, 25% and 10%, respectively). With even higher ARR frequencies (above 70%, 50% and 33%, respectively) the regime can be changed to severe selection.