The scientific community recognizes that molecular xenomonitoring (MX) can allow infected mosquitoes to serve as a proxy for human infection in vector-borne disease surveillance, but developing reliable MX systems for programmatic use has been challenging. The primary aim of this article is to examine the available evidence to recommend how MX can best be used for various purposes. Although much of the literature published within the last 20 years focuses on using MX for lymphatic filariasis elimination, a growing body of evidence supports its use in early warning systems for emerging infectious diseases (EIDs). An MX system design must consider the goal and target (e.g. diseases targeted for elimination versus EIDs), mosquito and pathogen characteristics, and context (e.g. setting and health system). MX is currently used as a ‘supplement’ to human surveillance and will not be considered as a ‘replacement’ until the correlation between pathogen-infection rates in human and mosquito populations is better understood. Establishing such relationships may not be feasible in elimination scenarios, due to increasingly dwindling human infection prevalence after successful control, but may still be possible for EIDs and in integrated disease surveillance systems. This article is part of the theme issue ‘Novel control strategies for mosquito-borne diseases'.