Two experiments were carried out in which low- and high-demand characteristics of self-reward were assessed in a multiple-baseline design across subjects. In the first experiment, arithmetic performance of four children was systematically assessed under no-reinforcement, self-reward/low-demand and self-reward/high-demand phases. In the second experiment, the performance of four children on a less meaningful task was assessed under the same conditions employed in Experiment 1. Results indicated that performance rates were relatively stable during the no-reinforcement phase but that these rates dropped markedly during the self-reward/low-demand phase for all eight subjects. Further, increased rates of performance were achieved under the self-reward/high-demand phase for all subjects. Results are discussed in terms of the assessment of the self-reinforcement procedure, current definitions of “self-reinforcement,” and the need to view self-reinforcement as existing on a continuum of external demand characteristics.