Colored Petri Nets (CPN) are a powerful, recognized and intuitive modelling tool. They allow a precise representation of distributed, embedded and/or real time systems. These models can be then used among others for the dependability assessment. This paper develops a new method of CPN analysis called the Backward reachability. It provides information about different ways of reaching a particular CPN marking that represents, for example, a failure state or a transilient fault. This analysis is performed on an inverse CPN which is obtained by transforming original CPN structure while preserving the model proprieties. The work develops mathematical tools to prove the pertinence of transformations allowing the definition of inverse CPN. The main advantage of this method is that it allows to determine the sequence leading from the initial to the final marking for any possible final marking vector.