Knowledge of wood frog distribution in Alaska is incomplete due to insufficient baseline occurrence data. A short season of activity and difficult access to remote areas restrict implementation of consistent monitoring efforts. Detecting the presence of species in aquatic landscapes using environmental DNA (eDNA) assays is increasingly applied as a monitoring method in wildlife surveys. However, uncertainties regarding the technique’s sensitivity to environmental variables and human error have thus far prevented its widespread adoption in studies of species distribution. Predictive models built on machine learning algorithms can help provide precise descriptions of species distribution using eDNA occurrence data, but they will require ground-truthing efforts to confirm accuracy in under-sampled landscapes. Here we assess the ability of wood frog eDNA occurrence data to inform species distribution models under five criteria for data use. We sampled 60 wetlands for eDNA in the Fairbanks North Star Borough during summer 2015. Samples were processed using a species-specific quantitative PCR assay. Wood frog presence at each site was inferred from the PCR results. This data was used to construct four different wood frog distribution models. From each model we produced a predictive distribution map encompassing the Fairbanks North Star Borough. We assess the performance of each model using available wood frog presence data. Our highest performing model achieves moderate predictive accuracy (Area Under the Curve = 0.74). Weak signals in eDNA occurrence data are important in revealing species presence at low abundance, but strict lab hygiene, quality control practices, and detailed metadata are needed to retain confidence in the results. We show a powerful new way to study wood frog distribution by combining eDNA occurrence data with machine learning techniques. Wider implementation of eDNA surveys and increased availability of high resolution GIS data will help to refine these models.