We used scale morphology to discriminate wild Atlantic striped bassMorone saxatilisfrom the Hudson River, Chesapeake Bay, and the Roanoke River. Morphological features used were perimeter shape as described by Fourier analysis, rectangularity, width and spacing of the first 10 circuli, and patterns of partitioning of the scale interior. Performance of discriminant functions was evaluated by examining the percent correct classification of known-origin samples (precision) and the bias in resulting stock composition estimates (accuracy). Correct classification rates ranged from 57% to 84%, varying with the number of stocks and year-classes included. Lowest rates were for three stocks with multiple year-classes; highest rates were for single-year-class analysis of the most geographically distant stocks. Despite some poor classification rates, bias and variance in stock composition estimates were low in four of five discrimination problems, suggesting that corrections based on the classification matrix could be applied with some confidence to unknown samples. However, we used bootstrapping to evaluate bias stability; further evaluation of independent samples would be useful.