The analysis of leaf physiognomy (size and shape) remains the most reliable means for reconstructing terrestrial paleotemperatures from before the Pleistocene. In particular, the strong correlation observed in living forests between the proportion of plant species that have untoothed leaf margins and mean annual temperature (MAT) is widely applied to fossil floras. Given its importance, it is striking that potentially more accurate methods based on more complete descriptions of leaf physiognomy are not considered reproducible. In an attempt to improve upon existing methods, we have created a protocol for analyzing leaf physiognomy using digital images and automated computer routines, called digital leaf physiognomy. We have developed procedures for processing digital images, including rules for selecting teeth for tooth area calculations. Critical rules include differentiating between primary and secondary teeth, and differentiating between teeth and lobes. Our procedures have already been readily adopted by multiple users, indicating that they are reproducible with only minimal training. The physiognomy of processed leaf images can be determined automatically using computer software: we focused on area-based variables such as tooth area, and perimeter-based variables such as perimeter ratio (perimeter / perimeter after teeth are removed). We applied these techniques to a transect of 17 present-day floras from the U.S. east coast and Panama; the standard error of MAT estimates for the best multiple linear regression is ± 1.9 °C, which is a significant improvement over the traditional univariate method involving presence vs. absence of teeth (± 3.1 °C). Digital leaf physiognomy shows promise for facilitating a new and improved class of paleoclimate methods. Moreover, these techniques are not restricted only to paleoclimate analysis, and also show considerable promise as proxies for the paleoecology of ancient trees and forests.