Avian colibacillosis is the main bacterial infectious disease in poultry and is caused by avian pathogenic Escherichia coli (APEC). However, E. coli strains are very diverse, and not all are pathogenic for poultry. A straightforward scheme for identifying APEC is crucial to better control avian colibacillosis. In this study, we combined high-throughput PCR and a machine learning procedure to identify relevant genetic markers associated with APEC. Markers related to phylogroup, serotype and 66 virulence factors were tested on a large number of E. coli strains isolated from environmental, faecal or colibacillosis lesion samples in 80 broiler flocks. Nine classification methods and a machine learning procedure were used to differentiate 170 strains presumed non-virulent (obtained from farm environments) from 203 strains presumed virulent (obtained from colibacillosis cases on chicken farms) and to develop a prediction model to evaluate the pathogenicity of isolates. The model was then validated on 14 isolates using a chick embryo lethality assay. The selected and validated model based on the bootstrap aggregating tree method relied on a scheme of 13 positive or negative markers associated with phylogroups (arpA), H4 antigen and virulence markers (aec4, ETT2.2, frzorf4,fyuA, iha, ireA, iroN, iutA1, papA, tsh, and vat). It had a specificity of 84 % and a sensitivity of 85 %, and was implemented as an online tool. Our scheme offers an easy evaluation of the virulence of avian E. coli isolates on the basis of the presence/absence of these 13 genetic markers, allowing for better control of avian colibacillosis.