Knowledge about the population size and trends of common bird species is crucial for setting conservation priorities and management actions. Multi-species large-scale monitoring schemes have often provided such estimates relying on extrapolation of relative abundances in particular habitats to large-scale areas. Here we show an alternative to inference-rich predictive models, proposing methods to deal with caveats of population size estimations in habitat-specialist species, reed warblers (Acrocephalus scirpaceus and Acrocephalus arundinaceus). Reed warblers were only found in pure reedbeds within riparian woodlands or in riparian vegetation scattered within or around reedbed patches, as expected according to their habitat specialization. The proportion of individuals located in reedbed associated with lotic and lentic waters differed between species, and no reed warbler was recorded in reedbed located along dry streams. This indicates that microhabitat features or their effects on reedbed structure and other factors made a proportion of the apparently available habitat unsuitable for both warbler species. Most warblers detected were males performing territorial singing (females seldom sing and do not perform elaborate territorial song, and are undistinguishable from males by plumage). The regional population sizes of the warbler species (~4000 individuals of A. scirpaceus and ~ 1000 individuals of A. arundinaceus) were much smaller than those estimated for the same area by transforming relative abundance obtained at a national scale to population size through extrapolation by habitat at a regional scale. These results highlight the importance of considering the habitat actually used and its suitability, the manner of sex-related detection, population sex-ratio and their interactions in population estimates. Ideally, the value of predictive methods to estimate population size of common species should be tested before conducting large-scale monitoring, rather than a posteriori. Although logistically challenging, this can be achieved by designing monitoring programs including an intensive sampling of abundance in ad hoc reference areas of variable size.