The use of Internet in the every day life has pushed its evolution in a very fast way. The heterogeneity of the equipments supporting its networks, as well as the different devices from which it can be accessed, have participated in increasing the complexity of understanding its global behavior and performance. In our study we propose a new method for studying the performance of TCP protocol based on causal graphs. Causal graphs offer models easy to interpret and use. They highlight the structural model of the system they represent and give us access to the causal dependences between the different parameters of the system. One of the major contribution of causal graphs is their ability to predict the effects of an intervention from observations made before this intervention.