Abstract: Working animals provide an essential transport resource in developing countries worldwide. Many of these animals are owned by poor people and work in harsh environments, so their welfare is a cause for concern. A protocol was developed to assess the welfare of working horses, mules and donkeys in urban and peri-urban areas, using direct observation of health and behaviour parameters. In this study, 4903 animals used for draught, pack and ridden work in Afghanistan, Egypt, India, Jordan and Pakistan were assessed between December 2002 and April 2003. The data showed that donkeys were more likely than mules or horses to demonstrate avoidance or aggressive behaviour towards an observer, while horses were most likely to make a friendly approach. Fewer than 8% of working equines had abnormal mucous membranes, ectoparasites or poor coat condition. Body lesions occurred predominantly in the areas of the breast/shoulder, withers and girth in all three species, with mules having the highest prevalence of lesions in these areas (22.5, 21.3 and 28.4%, respectively). Among horses and donkeys, the prevalence of these lesions was influenced by the type of work carried out. Lesions on the head, neck, ribs, flank and tail base were seen in less than 10% of animals. Across all three species approximately 70% of animals were thin, having a body condition score (BCS) of 2 or less on a scale of 1-5 (1, very thin; 5, very fat) and more horses were in very thin condition (BCS 1) than mules or donkeys. Over 75% of animals demonstrated limb deformities and abnormalities of gait. The results of this study are being used as the initial stage of a long-term strategy to inform priorities for welfare interventions in working equines and to establish a welfare benchmark. Subsequent stages will rank the welfare concerns identified, assess the contributing risk factors and implement specific interventions to address these risks. Following intervention, success in improving welfare will be measured by repetition of this protocol and comparison with the benchmark.