Abstract : Shear wave velocity (Vs) is an elastic parameter characterizing the stiffness of super- ficial formations. It is important both in applied seismology, for site effect or liquefaction estimation, and in civil engineering, for landslide investigation. This PhD work deals with the different seismic techniques available for deriving Vs values from ten to hundred meters depth. It aims at evaluating the capabilities of the different methods and defining a site characterization strategy adapted to their complexity. The first part of the work is devoted to the assessment of the surface wave inversion reliability for horizontally stratified medium. In collaboration with other European laboratories, active and passive seismic methods were applied at 20 European strong motion sites for which Vs profiles are known from borehole investigations. Comparison of results show that Vs profiles derived from surface (active and passive) seismics are similar for soft sites (Vs30 < 600 m/s) to those derived from borehole methods but present lower velocities for stiffer sites. Moreover, the information contained in dispersion curves appeared to be sufficient for EC8 site characterization. On the basis of the measured Vs profiles, a theoretical study on Rayleigh wave fundamental mode inversion was undertaken for sites with a strong contrast. A two step inversion procedure allowing bedrock depth estimation during the inversion process was defined on synthetic cases and successfully applied to horizontally layered real sites. More complex geometries are addressed in the second part. The inversion of dispersion curves obtained from active (synthetics and real data) and passive (synthetics) seismics over a dipping interface resulted in Vs profiles with a velocity gradient and misestimated bedrock depth and/or velocity. A procedure for detecting 2D/3D sites was defined from data of the 20 strong motion sites and of the Avignonet landlside (Isere, France). Based on active (seismogram, lateral spectrogram, dispersion curve comparison) and passive (shape of the H/V ratio, lateral variation of the H/V frequency peak) seismics, this procedure aims at adapting the choice of imaging technique during site characterization. In the third part, the ambient noise cross-correlation technique is applied to the Avignonet landslide. Thirteen seismological stations were settled on site during two weeks, and a tomographic inversion of arrival times measured with cross-correlation resulted in Rayleigh wave group velocity frequency maps. Their inversion allowed 3D Vs imaging of the landslide, down to 100 m depth. Finally, the evolution with time of ambient noise cross-correlations computed between two permanent stations on the landslide suggests a slight decrease in velocity over the three years of recordings. These results enlighten the usefulness of cross-correlation techniques for 3D investigation of smooth heterogeneous sites and for landslide monitoring.