Abstract : The growing number of musculoskeletal disorders in industry could be addressed by the use of collaborative robots, which allow the joint manipulation of objects by both a robot and a person. Designing these robots requires to assess the ergonomic benefit they offer. However there is a lack of adapted assessment methods in the literature. Many biomechanical quantities can represent the physical solicitations to which the worker is exposed, but their relevance strongly depends on the considered task. This paper presents a method to automatically select relevant ergonomic indicators for a given task to be performed with a collaborative robot. A virtual human simulation is used to estimate thirty indicators for varying human and robot features. A variance-based analysis is then conducted to extract the most discriminating indicators. The method is validated on several different tasks. The relevance of the proposed approach is confirmed by the obtained results.