1. Fish have been introduced into many previously fishless lakes throughout North America over the past 100+ years. It is difficult to determine the historical distribution of fishless lakes, however, because these introductions have not always been well-documented. 2. Due to its glacial history and low human population density, the state of Maine (U.S.A.) may host the greatest number of naturally fishless lakes in the northeastern United States. However, less than one-quarter of Maine's 6000+ lakes have been surveyed for fish presence, and no accurate assessments of either the historical or current abundance and distribution of fishless lakes exist. 3. We developed methods to assess the abundance and distribution of Maine's naturally fishless lakes (0.6-10.1 ha). We hypothesized that the historical distribution of fishless lakes across a landscape is controlled by geomorphic and geographic conditions. 4. We used ArcGIS to identify landscape-scale geomorphic and geographic factors (e.g. connectivity, surrounding slope) correlated with fish absence in two geomorphic regions of Maine - the western and interior mountains and the eastern lowlands and foothills. By using readily available geographic information systems data our method was not limited to field-visited sites. We estimated the likelihood that a particular lake is fishless with a stepwise logistic regression model developed for each region. 5. The absence of fish from western lakes is related to altitude (+), minimum percent slope in the 500 m buffer (+), maximum percent slope in the 500 m buffer (+) and percent cover of herbaceous-emergent wetland in 1000 m buffer (-). The absence of fish from eastern lakes is related to the lack of a stream within 50 m of the lake. 6. The models predict that a total of 4% (131) of study lakes in the two regions were historically fishless, with the eastern region hosting a greater proportion than the western region. 7. We verified the model predictions with two complementary approaches. First we visited 21 lakes predicted to be fishless and assessed current fish presence with gillnetting. Second, we used paleolimnological techniques based on the abundance of Chaoborus americanus mandibles in the bottom segments of sediment cores. Fifteen of the 21 lakes predicted to be fishless currently contain fish. Paleolimnological evidence, however, suggests that nine of the 15 lakes were historically fishless and thus were subject to undocumented fish introductions. 8. Our approach efficiently predicts the distribution Maine's naturally fishless lakes, and our results indicate that these habitats have declined due to fish introductions. Our method could be applied to other regions with similar geographic and geomorphic constraints on fish distributions as a tool to enhance conservation of a limited resource that provides habitat for unique biological communities. ?? 2007 Blackwell Publishing Ltd.