Nine-month-old infants' performance on discrimination-learning problems was investigated in four experiments using the synchronous reinforcement paradigm. These experiments were organized around basic theoretical postulates concerning the relation between attention and learning. In each of the experiments, infants were trained to respond differentially to a particular stimulus feature, with the goal of establishing whether they could learn to respond to a particular stimulus feature under conditions where other stimulus dimensions were present and varying. In the first experiment, 48 infants were trained to fixate visually on a particular feature in a pair of stimuli that varied in color, form, and position dimensions. Contingencies for responding were then shifted either within a dimension (reversal shift) or across dimensions (nonreversal shift). Infants learned to respond to the reinforced feature during initial training; moreover, infants assigned to a reversal shift condition showed a higher level of transfer of learning during the test phase than those assigned to a nonreversal shift condition. The second experiment extended the results of Experiment 1 by testing 48 additional infants under conditions in which the number of varying irrelevant dimensions was increased during the shift phase. Although the difficulty added to this task by this manipulation made transfer more difficult for all infants, results again indicated that transfer of learning was more evident for infants in a reversal shift than a nonreversal shift condition. In the third experiment, 64 infants were trained in a similar manner, except that completely new values were substituted during shift phases on the color and form dimensions. This manipulation was meant to probe whether infants were formulating a dimension response from previous training and to test the predictions of such dimension processing for transfer of learning to a functionally new problem. Infants were reinforced for fixating on a new feature either within the same dimension as during training (intradimension shift) or within the dimension that was not originally trained (extradimension shift). Transfer of learning was clearly superior in the intradimension shift condition. The findings of the first three experiments suggested that, during discrimination-learning problems, infants selected and tested individual stimulus features and dimensions from an array of potential solutions to the problem until discovering the one that was consistently associated with reinforcement.(ABSTRACT TRUNCATED AT 400 WORDS)