This thesis is dedicated to the probabilistic analysis of algorithms to reduce Euclidean networks. Euclidean network is all coefficients of linear combinations of a base integers (b_1, ..., b_n) \ subset R ^ n. The reduction of a network is to find a base formed of relatively short and relatively orthogonal vectors from a data base input. The famous LLL algorithm solves this problem efficiently in arbitrary dimension. It is widely used, but poorly understood. We focus on the analysis in the case n = 2, where LLL is the Gauss membership, as it is a building block for the case n> = 3, we analyze precisely the Gauss, both in of its execution (number of iterations, bit complexity, cost "additives") that the geometry of the output data (default ...