||The study is the first attempt to (i) model spring food webs in three SW Mediterranean ecosystems which are under different anthropogenic pressures and (ii) to project the consequence of this stress on their function. Linear inverse models were built using the Monte Carlo method coupled with Markov Chains to characterize the food-web status of the Lagoon, the Channel (inshore waters under high eutrophication and chemical contamination) and the Bay of Bizerte (offshore waters under less anthropogenic pressure). Ecological network analysis was used for the description of structural and functional properties of each food web and for inter-ecosystem comparisons. Our results showed that more carbon was produced by phytoplankton in the inshore waters (966–1234 mg C m-2 d-1) compared to the Bay (727 mg C m-2 d-1). The total ecosystem carbon inputs into the three food webs was supported by high primary production, which was mainly due to >10µm algae. However, the three carbon pathways were characterized by low detritivory and a high herbivory which was mainly assigned to protozooplankton. This latter was efficient in channelling biogenic carbon. In the Lagoon and the Channel, foods webs acted almost as a multivorous structure with a tendency towards herbivorous one, whereas in the Bay the herbivorous pathway was more dominant. Ecological indices revealed that the Lagoon and the Channel food webs/systems had high total system throughput and thus were more active than the Bay. The Bay food web, which had a high relative ascendency value, was more organized and specialized. This inter–ecosystem difference could be due to the varying levels of anthropogenic impact among sites. Indeed, the low value of Finn’s cycling index indicated that the three systems are disturbed, but the Lagoon and the Channel, with low average path lengths, appeared to be more stressed, as both sites have undergone higher chemical pollution and nutrient loading. This study shows that ecosystem models combined with ecological indices provide a powerful approach to detect change in environmental status and anthropogenic impacts.