This work aims to describe, implement and apply to real data some of the existing structure search methods with Bayesian Networks. Due to the vast dimensions of the graph space, complex search methods based on Markov Chain Monte Carlo (MCMC) are often required. In order to extract as much information as possible from the posterior distributions obtained with the MCMC methods, Bayesian model averaging is introduced and adapted to the particular case of Bayesian Networks. We apply the structure search methods to two different datasets. Firstly, we use a synthetic dataset whose graph is known a priori. This allows us to compare each of the search methods, as well as to check the convergence of the MCMC methods. Afterwards, we use a real dataset containing information about childhood neurodevelopment from a cohort of the INMA project. The three methods presented in this work have been implemented in R and C. The code has been made available as an R package on a public server at http://www.creal.cat/jrgonzalez/software.htm.. Bayesian Networks are graphical models that describe the probabilistic relationships between certain variables. They have been applied to a wide range of statistical issues. Among them is the discovery of biological structures such as disease-phenotype networks or gene-protein pathways. This is not an easy task because the number of possible graphs grows super-exponentially with the number of variables. As a result, complex search methods based on Markov Chain Monte Carlo (MCMC) are often required. The aims of this project are: analysing the statistical basis of these structure search methods, implementing them as efficiently as possible, comparing them using synthetic data and applying them to real biological data. For this last objective data of neurodevelopment during childhood will be used, in order to find the interdependencies between socioeconomical and cognitive-attention variables