1. Classical biological control (CBC) can be used to decrease the density of invasive species to below an acceptable ecological and economic threshold. Natural enemies specific to the invasive species are selected from its native range and released into the invaded range. This approach has drawbacks, despite the performance of specificity tests to ensure its safety, because the fundamental host range defined under controlled conditions does not represent the actual host range in natura, and these tests omit indirect interactions within community.2. We focus on Sonchus oleraceus (Asteraceae), a weed species originating from Western Palearctic that is invasive in many countries and notably in Australia. We explore how analyses of interaction network within its native range can be used to (a) inventory herbivores associated to the target plant, (b) characterize their ecological host ranges and (c) guide the selection of candidate biocontrol agents considering interactions with species from higher trophic levels. Arthropods were collected from plant community sympatric to S. oleraceus, in three bioclimatic regions, and interactions were inferred by a combination of molecular and morphological approaches.3. The networks reconstructed were structured in several trophic levels from basal species (plant community), to intermediate and top species (herbivorous arthropods and their natural enemies). The sub-network centred on S. oleraceus-related interactions contained 116 taxa and 213 interactions. We identified 47 herbivores feeding on S. oleraceus, 15 of which were specific to the target species. Some discrepancies with respect to published findings or conventional specificity tests suggested possible insufficient sampling effort for the recording of interactions or the existence of cryptic species. Among potential candidate agents, six exhibited interactions with natural enemies.4. Synthesis and applications. Adopting a network approach as prerequisite step of the classical biological control programme can provide a rapid screening of potential agents to be tested in priority. Once ecological host range defined, we suggest that priority should be given to agent used by a minimum species, and, when they