The network build-up of epoxy-anhydride thermosets has been studied using two different stochastic network build-up methods based on the random combination of primary chains or simple fragments. Explicit expressions for relevant statistic averages in the pregel and postgel states have been derived for both methods. The application of both methods to a living polymerization kinetic model leads to strong divergences in the vicinity of the critical conversion at gelation due to growing differences between the real primary chain distribution and the most probable distribution in the fragment model. However, the application of both methods to an initiator regeneration kinetic model leads to identical results because of the distribution of primary chains throughout the whole curing process is a most probable one. (C) 2014 Elsevier Ltd. All rights reserved.