Microtremors are increasingly used in earthquake engineering to infer the shear-wave velocity profile at a given site. Assuming they are mainly composed of surface waves, ambient vibrations recorded by an array of sensors can be used to determine the dispersion curve. Generally, it provides a large frequency band dispersion curve and it has the advantage of not requiring artificial sources, making it particular suitable for urban applications. Due to the data uncertainties and the non-linearity of the problem, the solution of the dispersion curve inversion is not unique. Direct search methods like the neighbourhood algorithm allow the investigation of the whole parameter space and the introduction of prior information in a rational way. Due to the limited number of parameters in surface-wave inversion, they constitute an attractive alternative to linearized methods. During this thesis, efficient tools based on the neighbourhood algorithm are developed to obtain the one-dimensional 
profile from passive or active source experiments. As the number of generated models is usually high with stochastic techniques, special attention is paid to the optimization and to the reliability of the forward computations.
The developed code has been tested on several synthetic models, among them one is presented here. The effects of the available frequency range and the influence of the prior information are particularly emphasized. Higher modes might bring additional constraints during the inversion but they also raise the crucial problem of their correct identification, for which an algorithm is proposed. We also show that the inversion of Love and Rayleigh modes is a promising technique to increase the penetration depth of the method. Moreover, we developed a specific tool for the inversion of auto-correlation curves which takes into account the uncertainties observed on experimental curves and propagates it to the inverted velocity profiles.
The whole interpretation chain from field acquisitions to the achievement of velocity profiles is illustrated by two examples with synthetic and real wavefields (Liège, Belgium). Information from boreholes, classical refraction tests, active surface wave experiments, and from the H/V peak frequency are analysed to check the validity of the array results.