The computations of fuel tanks are bound to cover uncertainties related to geometric and material imperfections, post-welding stresses, non-uniform settlement, and shell degradation due to corrosion. The paper compares four different methods of tank corrosion description: a uniform reduction of the sheet thickness of the entire shell, degradation described by an angle correlated with its partial fuel filling, corrosion patterns defined by appropriately selected trigonometric functions, and an advanced non-homogeneous model using theoretical random fields. All corrosion patterns were numerically investigated to identify their impact on structural response. The computations were carried out for a simplified numerical model of a mounded horizontal pressure vessel. The Point Estimate Method (PEM) was used to estimate the mean value and standard deviation of the shell critical force. The probabilistic approach allows to assess structural reliability and makes it possible to optimize the structure. It has been shown that the optimal variant of corrosion description, easy for engineering applications, is the uniform reduction of the shell thickness.