Even though the saturation in water of most natural and engineered soils is partial, constitutive models in geotechnical engineering have long made the assumption of complete saturation. The increasing knowledge of the rheological behaviour of unsaturated soils is now favourable to new constitutive formulations to understand the stress-strain behaviour under the effect of suction. Yet, it is not agreed that there is a unique way to account for capillary effects, and several families of constitutive models appear to be either insufficient to describe the complete soil behaviour or else far too complex to be handled in practical engineering. The principal objective of the PhD work is to overcome those limitations by formulating a new unified and advanced constitutive model to understand and predict the behaviour of unsaturated soils. It is specified that the model shall be applicable to the broadest panel of civil engineering cases involving changes in water content of the soil, for instance landslides and earthdams. The methods and contributions of the presented work can be summarized in four main themes that are (i) the effective stress (ii) the constitutive model for stress-strain behaviour, (iii) the model for water retention behaviour (iv) the integration of the coupled model in a finite element code. A review and clarification of the possible effective stresses for unsaturated soils is proposed. The generalized effective stress is identified and justified as the most suitable and convenient unique mechanical effective stress for stress-strain behaviour. It combines the mechanical stress and the capillary variables that are degree of saturation and matric suction. The complete stress-strain framework is built on the basis of the generalized effective stress by relying on work input equation. A new constitutive model named ACMEG-s is formulated from a reference critical state elasto-plastic model which features two mechanisms of plasticity. The proposed model uses bounding plasticity. The effects of capillarity are introduced by the means of a modified yield limit, and the internal couplings due to the effective stress itself. The mechanical behaviour is thus dependent on the matric suction and degree of saturation. The model shows particularly good performances in modelling wetting pore collapse and swelling pressure tests. On the basis of experimental results obtained with an unsaturated oedometer, new contributions to the understanding of the soil water retention curve were produced. A new model linking the degree of saturation and the matric suction has been formulated and validated on a number of experimental datasets. An innovative interpretation of the coupling between the void ratio and air entry value is proposed, in parallel with discussions on the capillary hysteresis. The retention model, based on kinematic hardening, is thus added to the stress-strain model which requires the management of double-way coupling. The coupled model ACMEG-s has been implemented into the finite element code LAGAMINE, in the objective of application to boundary value problems. The implementation required the development of specific integration routines for the constitutive models with a constant updating of the terms of coupling. The numerical code enables a three-level coupling between mechanical, retention, and fluid flow behaviours. Two case studies are presented with innovative insight into the spatial distribution of pore water pressures: the earth dam of Mirgenbach and the shallow landslide experiment of Rüdlingen. Recommendations are also formulated with respect to parameter determination.