Abstract Mosquito surveillance programmes are essential to assess the risks of local vector-borne disease outbreaks as well as for early detection of mosquito invasion events. Surveys are usually performed with traditional sampling tools (i.e., ovitraps and dipping method for immature stages or light or decoy traps for adults). Over the past decade, numerous studies have highlighted that environmental DNA (eDNA) sampling can enhance invertebrate species detection and provide community composition metrics. However, the usefulness of eDNA for detection of mosquito species has, to date, been largely neglected. Here, we sampled water from potential larval breeding sites along a gradient of anthropogenic perturbations, from the core of an oil palm plantation to the rainforest on São Tomé Island (Gulf of Guinea, Africa). We showed that (i) species of mosquitoes could be detected via metabarcoding mostly when larvae were visible, (ii) larvae species richness was greater using eDNA than visual identification and (iii) new mosquito species were also detected by the eDNA approach. We provide a critical discussion of the pros and cons of eDNA metabarcoding for monitoring mosquito species diversity and recommendations for future research directions that could facilitate the adoption of eDNA as a tool for assessing insect vector communities.