The aim of this work is to handle simplified models of carbonation in a probablistic approach in order to propose an optimized maintenance strategy against steel corrosion. First of all, a synthesis of existing simplified models is presented. Three categories of models have been highlighted. Then a sensitivity analysis is proposed for a ranking of input parameters. Elasticity, linear correlation coefficient of Pearson, impact on the mean and the standard deviation of the models response are the indicators used. Given that corrosion concerns overall a structure surface, a methodology to take into account spatial variability through these models is described in this paper. The method used for describing spatial variability is Karhunen Loeve expansion. When inspections are carried out on the strucure, new data are available. Therefore, an updating process of the spatial variability parameters through bayesian network has been tested in this work. Results show that the updating process works correctly.