The main objective of a typical identification process is to determine the best input parameter set of a given problem for which the cost function is not explicitly known in terms of its associated inputs. In the worst case, the cost function is largely multimodal and needs an efficient identification technique to reach an acceptable solution within a reasonable duration. In this work, we propose a new identification hybrid strategy based on genetic algorithm with parallel selection (GAPS) and artificial neural network (ANN). This strategy is used as an identification scheme in the modelling of non-linear mechanical behaviour of laminate composites. ANN is designed in a way to relate the mechanical parameters of the studied material to the cost function representing the difference between the true and simulated mechanical responses. The identification problem assumes different mechanical behaviours related to non-linear model of composite laminate shell including elasticity, plasticity, viscoelasticity, viscoelasticity and damage. The strategy is built based on the following idea: because GAPS is a time consuming technique with regard to gradient based methods, ANN is used as a meta model to speed up the identification process while GAPS provides the required database from early generations for ANN training. The hybrid strategy is then able to solve the identification problem as demonstrated by the well agreement between experimental and simulated mechanical tests (i.e., traction and internal pressure tests) thanks to the robust character of GAPS and the rapid convergence of ANN.