Our aim was to compare the efficiency of different surveillance strategies for detecting scrapie-infected sheep flocks in the Norwegian population using simulation modelling. The dynamic Monte Carlo simulation model has the flock as the unit. The input parameters include properties of the sheep population (number of flocks, flock size, age distribution, reasons for culling, breeds, prion protein-allele distribution); properties of scrapie (genotype-dependent infection rate and incubation periods, and age- and genotype-dependent prevalence of scrapie); properties of the surveillance strategy (selection of sheep for examination, period in which infected sheep are detectable, and properties of the diagnostic tests). For simplification, the prion protein-alleles were grouped into three allele groups: VRQ, ARR, and ARQ' (ARQ' represents ARQ, ARH and AHQ). Through either abattoir surveillance or surveillance of fallen stock, <or=9% of scrapie flocks were detected. The necessary sample size for detecting any particular number of scrapie flocks was considerably lower using surveillance of fallen stock than abattoir surveillance. After increasing the diagnostic method's sensitivity, only the efficiency of abattoir surveillance increased. The prion protein-genotypes ARQ'/ARQ', VRQ/ARQ' and VRQ/VRQ were overrepresented both in the sampled infected sheep and in the detected sheep. Sheep with ARQ'/ARQ' constituted >70% of the detected sheep (compared to 33% in the underlying population). The model output was sensitive to the susceptibility of infection for the genotype ARQ'/ARQ'. The effect was large for abattoir surveillance (increased susceptibility increased the efficiency of abattoir surveillance).