Résumé : A time-consuming problem encountered both in system diagnosis and planning is that of computing trajectories over a behavioral model. In order to improve the efficiency of this task, there is currently a great interest in using model-checking techniques developed within the area of computer aided verification. In this paper, we propose to represent the system as automata and we define a property called inversibility. This property is used to improve the efficiency of the search algorithm computing trajectories. We present two study cases in diagnosis and planning domains where this approach gives satisfactory results.