The chemical computing model was originally proposed as a simple and elegant parallel programming paradigm. Data are seen as molecules and computation as chemical reactions : if some molecules sat- isfy a predefined reaction condition, they are replaced by the product of the reaction. When no reaction is possible, a normal form is reached and the program terminates. In this paper, we describe how we can build fault tolerant autonomic systems in a chemical setting. We pay a partic- ular attention to the chemical description of the simple and successful constructs for fault-tolerance such as the recovery block scheme by de- signing a generic chemical framework. Then, we apply this framework to the development of an autonomic mailing system... all this in a chemical setting.