Bone is a natural cellular composite, with a gradient structure going from a loose interconnected core to an outer dense wall, thus minimising bone weight while maintaining a high mechanical performance. Bone can be repaired using either auto- or allo- grafts, which, however, have a limited availability, or present pathogen transmission risks, respectively. Synthetic tissue grafts are therefore investigated in order to provide porous scaffolds, seeded with the appropriate type of cells, as templates for tissue regeneration. This study focused on composite scaffolds based on polymer matrices. The objectives were to understand how different foaming phenomena are affected by the addition of fillers into the polymer, to investigate their effect on the structural and mechanical properties of foams, and then to propose and optimise a solvent free process in order to prepare bioresorbable composite scaffolds suitable for bone tissue engineering applications. The comparison of several processes resulting in porous matrices led to choose supercritical CO2 foaming as a flexible technique to manufacture cellular structures, with the possibility of adding fibres or fillers. First, the integration of glass fibres into polyurethane (PUR) and polyethylene terephtalate (PET) foams, processed by reactive and gas foaming respectively, was investigated in order to study the potential improvement of mechanical properties of porous polymers. With the PUR system, modulus was shown to increase with fibre content, and an optimum fibre length existed. With PET, dense and heterogeneous structures were obtained. In both cases a homogeneous mixing of the two components was required before foaming. Gas foaming of bioresorbable neat polymers, poly(L-lactic acid) (PLA) and poly(D,L-lactic-co-glycolic acid) (PLGA), was carried out in a high-pressure vessel. A wide variety of PLA cellular structures was obtained, with open and closed pores, of diameters between 0.2 and 1.0 mm, and a range of porosity (70-92 %) and modulus (10-180 MPa). On the contrary, using PLGA led to closed, larger pores, and a low modulus. Only PLA was therefore considered as a matrix for composite foaming. Hydroxyapatite (HA) and β-tricalcium phosphate (β-TCP), were then added to PLA in order to improve mechanical resistance, to regulate pH during scaffold resorption, and, owing to their intrinsic osteoconductivity, to favour mineralisation. Composite preforms were prepared by three different techniques: mixing powders in the dry state, dispersion in polymer solution, and melt-extrusion. Observation of the ceramic distribution, measurements of glass transition and melting temperatures, and tensile tests led to select and optimise melt-extrusion as a clean technique for preparing homogeneous ceramic-polymer preforms. Structural analysis of composite foams produced in the highpressure chamber enabled us to define processing windows as a function of pressure release and cooling rates. HA or β-TCP filled PLA foams were obtained, with porosity between 74 and 83 %, pore size between 0.1 and 1 mm, and modulus between 30 and 250 MPa. Microcomputed tomography confirmed the anisotropy in cellular morphology. Compression tests were also performed and revealed the anisotropic and viscoelastic behaviour of the porous polymer and composite structures. Finally, in collaboration with the Laboratory of Orthopaedic Research (LRO-EPFL) and the Hôpital Orthopédique de Suisse Romande (HO-CHUV), the biocompatibility of the developed structures with human primary foetal bone cells was demonstrated, as they proliferated on the template surfaces and differentiated. Foaming parameters could thus be defined to create composite scaffolds suitable for bone tissue engineering applications. Structural and mechanical properties similar to those of cancellous bone were demonstrated and promising in vitro tests were conducted. Therefore the developed scaffolds can now be evaluated in vivo.