Abstract: An evolutionary neural network constitutive model is proposed to describe the whole stress-strain process of rock under chemical corrosion. Differing from other conventional models, the model reflects the whole stress-strain process of rock under chemical corrosion in terms of its network topology and weights and it is constituted by genetic algorithm with favorable prediction function. After training with samples in which sufficient stress-strain information are involved, the model will indicate the complete stress-strain relationship of a mineral that is the same to the sample which has been corroded by different chemical solutions. This has been proved by a practical example. The model is recommended to apply for prediction of the whole stress-strain process of a rock that would not participate in the training by evolutionary neural network.