Identifying sex of hatchling turtles is difficult because juveniles are not obviously externally dimorphic, and current techniques to identify sex are often logistically unfeasible for field studies. We demonstrate a widely applicable and inexpensive alternative to detect subtle but significant sexual dimorphism in hatchlings, using landmark-based geometric morphometric methods. With this approach, carapace landmarks were digitized from photographs of each hatchling, and shape variables were generated after variation in size, location and orientation were eliminated. These variables were then analyzed for sexual dimorphism, and used in discriminant function analysis to estimate sex of each hatchling. Using this approach on two species (Chrysemys picta and Podocnemis expansa), we found this method had high accuracy in assigning sex when compared with true sex (98% and 90%, respectively), and cross-validation revealed a correct classification rate of 85%. These correct classification rates were considerably higher than those found on the same species using linear distance measurements as data. We also explored two alternative statis- tical approaches for assessing sex (K-means clustering and multiple logistic regres- sion) and found that these alternative approaches were accurate only 61% and 78% of the time, respectively, in C. picta and 69% and 77% of the time in P. expansa. These findings are similar to classification rates found for turtle species using ap- proaches based on linear distance measurements. We also found that the observed sexual dimorphism differed between the two species. In P. expansa, males displayed relatively more expansion of the central region of the carapace relative to females, whereas in C. picta this pattern was reversed. We conclude that discriminant analysis of morphology quantified using geometric morphometrics provides researchers with a powerful tool to discriminate between male and female hatchling turtles.