This paper presents a 3D reconstruction method of the proximal femur shape based on contour identification from pairs of 2D X-ray radiographs. The aim is to reconstruct the 3D surface of the proximal femur from a limited number of projections. Our approach is based on the reconstruction of several 3D contours, which are meshed to obtain a 3D shape. Our proposed algorithm is based on different processing steps to obtain the 3D personalized model. Three approaches are proposed. The first technique consists of contour extraction, matching the points of these contours and calculation of a 3D contour using an original algorithm. The second and third techniques use the results of the previous method as well as new points chosen by an operator from a reference model of femur to improve accuracy either globally or only in sensitive areas. The proposed method was evaluated on 10 cadaveric proximal femurs using the 3D CT-Scan models. Obtained results show good performance and promising perspectives for 3D shape reconstruction of the femur from only a few pairs of radiographs.