Digital elevation models (DEMs) have been found to be an effective data source for automated mapping of wetlands. However, it is unclear whether high spatial resolution DEMs, which tend to be more expensive to acquire and process, are necessary for mapping wetlands such as those in the US National Wetland Inventory (NWI). Therefore, we compared predictions of the probability of palustrine wetland occurrence with a random forests (RF) algorithm using DEMs generated from light detection and ranging (LiDAR) at 1 m, 3 m, and 10 m raster cell sizes; and photogrammetrically-derived DEMs at 3 m and 10 m. For each classification, a wide range of terrain derivatives were generated and used as the input data for the classification. Comparisons between the wetland predictions were made using the receiver operating characteristic (ROC) area under the curve (AUC) measure, the Kappa statistic, overall accuracy, class user’s and producer’s accuracy, and the out of bag (OOB) error rate. For two different study sites, irrespective of the source of the digital terrain data, palustrine wetland occurrence was predicted with AUC values greater than 0.95, overall accuracies greater than 88%, Kappa greater than 0.77, and wetland user’s and producer’s accuracies above 0.85 when using a large training data set derived from the NWI or a small separate data set of non-NWI data derived from field samples. We therefore conclude that the source (LiDAR vs photogrammetric) and spatial scale (1 m, 3 m, or 10 m) of the DEM data does not have a large impact on the accuracy of the prediction of wetlands such as those in the NWI. However, for small wetlands, or more generally for wetlands unlike those in the NWI, finer scale data (e.g. 1 m) derived from LiDAR may be preferable.