Training operators of complex interactive systems is a difficult task that involves multiple actors, requires specific competencies and may be extremely costly in terms of time and resources. For instance, an initial training has to be performed in order to bring operators to a desired level of declarative and procedural knowledge. A successful operator will be granted a qualification to operate the system that was used for (or targeted by) the training. However, depending on the nature and the diversity of the operations performed on a daily basis, this initial training may decay and some knowledge may become deprecated. Recurrent training needs to be organised on a regular basis in order to keep operators qualified for the work. Such training differs from initial training and requires taking into account information about how humans learn and forget, what has been performed by operators since last training and the expected required level of knowledge. Trainers need to encompass all this generic information (to all operators) and specific information (to each individual) to define both initial and recurrent training programs. One particularly cumbersome task is the gathering of what happened during operations for each operator in order to minimize recurrent training program to knowledge that might have been forgotten as it was not used in recent operations. We propose a tool-supported task-model approach fed by information of the real work of operators in order to identify complete, relevant and efficient training for operators. These contributions have been applied to the civil aviation domain demonstrating multiple induced benefits.