The dynamics of a neuron are influenced by the connections with the network where it lies. Recorded spike trains exhibit patterns due to the interactions between neurons. However, the structure of the network is not known. A challenging task is to investigate it from the analysis of simultaneously recorded spike trains. We develop a non-parametric method based on copulas, that we apply to simulated data according to different bivariate Leaky In- tegrate and Fire models. The method discerns dependencies determined by the surround- ing network, from those determined by direct interactions between the two neurons. Furthermore, the method recognizes the presence of delays in the spike propagation.