We present a novel hierarchical multiscale methodology aimed at generating the atomistic structure of epoxy networks and predicting their properties and topology. Our approach combines several computational techniques and consists of four separate procedures: (1) mapping of the polymerizing monomers onto a coarse-grained representation using the method of neural-gas networks, (2) cross-linking the coarse-grained monomers at mesoscale by applying dissipative particle dynamics and a probabilistic reaction scheme, (3) reverse mapping of the coarse-grained polymer network to a fully atomistic representation, and (4) simulation of the atomistic model by means of molecular dynamics technique. As a case study, we simulate DGEBA + DETDA epoxy network formation and analyze the thermal properties and the network topology. It was found that the parameters used for generating the atomistic structures (degree of coarse graining in our case) can significantly influence the network topology and properties. Moreover, we show that rather big simulation boxes are necessary to obtain the proper local structure of an epoxy resin.