A computational methodology is presented for modelling the material non-linear mechanical behaviour of composite structures made of FRP (Fibre-Reinforced Plastic) laminates. The model is based on the appropriate combination of the constitutive models of component materials, considered to behave as isolated continua, together with additional ‘closure equations’ that characterize the micromechanics of the composite from a morphological point of view. To this end, any appropriate constitutive model may be selected for each phase. Each component is modelled separately and the global response is obtained by assembling all contributions taking into account the interactions between components in a general phenomenological way. To model the behaviour of a single unidirectional (UD) composite lamina, a Serial-Parallel continuum approach has been developed assuming that components behave as parallel materials in the fibres alignment direction and as serial materials in orthogonal directions. Taking into account the internal morphology of the composite material, it is devised a strategy for decoupling and coupling component phases. This methodology [1], named “compounding of behaviours”, allows to take into consideration local degradation phenomena (in the constituents materials), like plasticity, damage, fatigue [2], etc. in a coupled manner. It is based on the proper management of homogenous constitutive models, already available for each component. In this way, it is used all developments achieved in constitutive modelling of plain materials, what makes possible the transference of this technology to composites. A lamination theory complemented with the proposed UD model is employed to describe the mechanical behaviour of multidirectional laminates. A specific solution strategy for the general non linear case is proposed. It provides quick local and global convergences, what makes the model suitable for large scale structures. The model brings answers on the non-linear behaviour of composites, where classical micro-mechanics formulas are restricted to their linear elastic part. The methodology is validated through several numerical analyses and contrasted against experimental data and benchmark tests.