Full scale simulators have introduced new approaches to the study of the subject's ways of acting during problem situations, allowing prospective and repeated observations. Using this technique, we compared and analyzed the responses of anesthesia trainees with more or less experience to 4 simulated problem situations in order to better understand the nature of expertise in dynamic problem situations, what makes problems more or less difficult, and prevent human error. The incidents were chosen for their variable complexity ( familiarity, interrelation of indices, time characteristics). Data were collected from systematic observations and videotaping of simulation sessions and self-confrontation. Quantitative and qualitative variables were used to compare the performance of subjects: accuracy and speed of diagnosis, type of actions, identification and selectivity of diagnostic indices, sequence of hypotheses generated, type of reasoning. From the analyses of the results, a model of problem solving in dynamic situations emerges across incidents. Hypothesis were formed about the type of knowledge and strategies that are lacking in novice and how it could possibly be improved. At the same time, we raised the issues concerning the validity of the use of simulators to study performance and propose some methodological elements.