The effectiveness of three tutorial strategies was evaluated by experimentally manipulating the strategies used in a 2 X 3 factorial design. Two tutors taught 48 college undergraduates a basic optics lesson about "how lenses work," using three instructional methods which varied the amount of tutor control. In the lecture condition, tutors assumed full control; in the discovery condition, students assumed full control; and in the contingent condition, tutor control was adapted to the needs of individual students. Serving as a baseline, eight students learned the same lesson from a text. Generalized regression analysis was used to assess the effects of aptitudes, treatments, and their interactions (ATI). Significant ATI suggested that discovery methods were particularly ineffective with low ability students, whereas less able students benefited from contingent techniques, and lecture techniques produced intermediate results. When correctly applied, contingent instruction produced the highest learning outcomes across the ability range; however, this strategy proved difficult to carry out consistently. Implications for tutor training and the development of "intelligent" tutoring systems are discussed. Contains 21 references. (Author)