Quasars are among the most luminous and the most distant objects in the Universe. Consequently they are particularly interesting to probe its origin and to understand its evolution. However, the huge distances at which these objects are generally found prevent us from resolving their central regions so that we cannot directly check the validity of the geometrical as well as the dynamical models accounting for their observational properties (spectral energy distribution, line profiles, presence or absence of radio jets etc). In our thesis, we use two indirect observational techniques in order to constrain the existing models. These techniques which are particularly sensitive to the geometrical structure of the quasar emission regions are polarimetry and gravitational microlensing. In the first part of our thesis we study the correlation between the direction of the linear polarization and the orientation of the host galaxy/ extended emission that we determined on the basis of high resolution HST images. We show how this study enables us to bring new clues favoring the existence of a unification model for the Type 1 and Type 2 quasars. In the second part, we show how gravitational microlensing allows to constrain the geometry and size of the regions at the origin of the broad absorption lines observed in the spectrum of 10 to 20 % of quasars. For this purpose we build a radiative transfer code allowing to simulate the line profiles produced in a variety of realistic wind models. These models are then used to study the variations of line profiles induced by the transit of a gravitational microlens. This technique is finally applied to the case of the quasar H1413+117 in order to determine the geometry of the regions which produce the broad absorption lines.