Two studies were conducted to clarify the role of different minimum performance standards for contingent tangible reinforcement or self-dispensed evaluative reinforcement (in the absence of tangible rewards) in determining the rate and accuracy of learning. Preschool children were presented with a discrimination learning task. Their performance had to meet low (few correct), medium, or high (all correct) standards to be rewarded. In an accelerating standard condition their performance had to surpass that on the previous trial. In one experiment, rewards were externally dispensed tokens, while in the second, children self-administered verbal evaluations ("I'm doing real good!"). Learning was significantly more rapid in the high and accelerating standards, and the self-dispensation of evaluative reinforcement produced such rapid learning that by the end of the experiment, effects due to different standards had vanished. Results are discussed in terms of the incentive values of differing standards and the power of self-dispensed evaluative reinforcement. (Author/BRT)