This report proposes a CAD-based approach for identification of the elasto-static parameters of the robotic manipulators. The main contributions are in the areas of virtual experiment planning and algorithmic data processing, which allows to obtain the stiffness matrix with required accuracy. In contrast to previous works, the developed technique operates with the deflection field produced by virtual experiments in a CAD environment. The proposed approach provides high identification accuracy (about 0.1% for the stiffness matrix element) and is able to take into account the real shape of the link, coupling between rotational/translational deflections and joint particularities. To compute the stiffness matrix, the numerical technique has been developed, and some recommendations for optimal settings of the virtual experiments are given. In order to minimize the identification errors, the statistical data processing technique was applied. The advantages of the developed approach have been confirmed by case studies dealing with the links of parallel manipulator of the Orthoglide family, for which the identification errors have been reduced to 0.1%