Researchers in many disciplines face the formidable task of analyzing massive amounts of high-dimensional and highly-structured data, and most of the time, they are in the situation of processing data whose generative source is uncertain. Several models and techniques have been developed to satisfy the necessities of this huge task, and most of those techniques need to define different parameters that should be chosen for the good behavior of estimation or classification of future data. In this thesis we focus in the important, and highly used L1-regularization, and how to choose its corresponding regularization parameter for Generalized Linear Models (GLM) by using a Quantile Universal Threshold (QUT) method, and in this way we show an automatic method to obtain the vector of coefficients and therefore, to be able to perform estimation and model selection in GLM. We show some experimental results that corroborate the capacity of QUT to include the true models in simulated data, and some results with real data.