Summary Musculoskeletal disorders during manual handling (MH) are still very common in our workplaces. The biomechanical risk factors most often associated with back pain include manual handling, excessive repetitions, trunk rotation and lifting of heavy loads. Several determinants of an MH task were studied to determine their importance for workers’ physical exposure. For example, the height of the load, the initial distance of the load from the body, the mass lifted and the lifting speed are all determinants related to the magnitude of the external moment at L5/S1 or of the compression force. One of the most frequently promoted preventive measures in MH promotes lifting boxes in a symmetrical posture in order to avoid trunk rotation movements. One way to limit asymmetrical postures consists in facing the load to be lifted and letting the feet move freely. Foot placement is a key parameter to describe handlers’ motor behaviour as they approach the load to be lifted, decrease asymmetrical postures, and move from the pickup site to the deposit site (transition phase). However, there is very little information available on the various foot movement strategies applied during this phase. One study has proposed metrics to classify and quantify foot movements (Wagner et al., 2009, 2010), but that method had significant limitations. The general objective of this study was mainly to appropriate (Wagner, Kirschweng et Reed, 2009; Wagner, Reed et Chaffin, 2010) method and adapt it to the needs of this study and to validate the improved method by quantifying handlers’ movements. A new taxonomy that can be used to determine handlers’ foot movement strategies was therefore developed and validated, then applied to the existing data from Plamondon et al. (2010) et de Plamondon et al. (2014) to define the most common foot placement strategies (chapter 3) and strategies used by expert and novice handlers (chapter 4). Since machine learning techniques are becoming increasingly popular and reduce analysis time (especially by limiting manual observation time), a technique involving automatic classification of foot placement strategies by means of machine learning was set out (chapter 5), and then tested by comparing actual observations with those predicted by this technique (chapter 6). Finally, an experimental phase in the laboratory involved applying the step detection method to 15 novice handlers under the effects of four key determinants in handling: the pickup and deposit heights; the transfer distance (which constrains the possibility of moving the feet); the mass of the load; and the pace (chapter 7). Among major benefits, we now have a method to classify foot placements by means of observation and automatically. New knowledge of experts’ and novices’ foot placement and movement strategies has also been added and will make it possible to enhance training programs for handlers. For example, experts tend to opt for a more static and gradual foot movement strategy, whereas novices’ movements are more fluid and variable. Foot placement strategies also affects the resulting moments and asymmetrical movements at the time of lifting. Finally, this improved method should allow better documentation of handlers’ foot movements in the laboratory and, eventually, during on-site observations of real-life work.