This randomized study examined the effectiveness of a preschool stimulation program created to teach words that had been selected by considering the needs of the target population of children. Twenty-two educators and their group of at-risk preschoolers (N=222, M age=4.27 years) were assigned to one of two conditions: control or intervention. In the latter condition, educators had to read specifically developed storybooks to their group and conduct stimulation activities. Despite the training and support they received, educators implemented the intervention with varying degrees of fidelity. Nonetheless, intent-to-treat comparison of the two conditions indicates that children in the intervention condition learned the meaning of a much greater number of words than their peers in the control condition. In addition, efficacy subset analyses that took into account fidelity of implementation show that the greatest gains were made by children who had an educator who had implemented the intervention reliably. Strategies for scaling up the intervention and optimizing its implementation are discussed. (author abstract)