African swine fever (ASF) is present in most of Sub Saharan Africa, Sardinia, Transcaucasia and Russia. Most outbreaks affect backyard and small-scale farms (SSF), and transmission between farms involves movements of infected pigs or products. Field studies show that in case of a suspected outbreak, farmers often sell pigs with clinical signs for slaughter and pigs without signs to traders. ASF's incubation and latent periods and farmers' imperfect clinical diagnosis (sensitivity [Se] and specificity [Sp], time to detection and sale of animals [t]) facilitate the risk of selling infected pigs without detected signs to traders. This study aimed at estimating the probability that SSF release ASF infected pigs without detected signs via emergency sale and the impact of improving farmers' diagnosis. A stochastic, individual-based, discrete time, state-transition model simulating ASF spread within SSF and emergency sale was developed in R. The model was parameterised using results from field surveys in Africa and the literature. The probability of releasing infected pigs without detected signs (Pr), the proportion of the herd released and the effect of t, Se and Sp were assessed for SFF of different sizes. For SSF of 10 pigs and more, t, Se and Sp of farmers' clinical diagnosis have little effect on Pr, which remained high under all scenarios (from Pr=0.33 [n=10, t=35, Se=0.9, Sp=0.75] to Pr=0.93 [n=30, t=25, Se=0.75, Sp=0.9]). Although very small herds (n<10) constitute the majority of herds in affected regions, larger SFF (10<n<100) contributed proportionally more to ASF persistence via emergency sale. Improving farmers' diagnostic accuracy is not an effective mitigation strategy as it does not significantly reduce the release of ASF infected animals via emergency sale. Alternative control measures involving other stakeholders need to be explored, such as market-based or certification approaches. (Texte intégral)