Despite a large body of research demonstrating that generalization to novel stimuli can be produced by training sufficient exemplars, the methods by which exemplars can be trained remain unclear. The purpose of the current study was to evaluate 2 methods, serial and concurrent presentation of stimuli, to train sufficient exemplars. Five preschool children with developmental delays were taught to identify letters or letter sounds using serial and concurrent presentation. Generalization to untrained exemplars was evaluated for targets trained using each method. Participants reached the mastery criterion in fewer training sessions, on average, using the concurrent method of presentation than the serial method, and the concurrent method also resulted in greater generalization to untrained exemplars.