Experiment I was designed to demonstrate that young children fail to abstract the positive cue as the relevant stimulus event in a restricted concept-learning task. Sixteen kindergarten and 16 fourth grade subjects were trained to criterion on a Kendler-type task, whereupon each subject was presented a pair of new instances which contrasted only in the original cues of the original relevant dimension. Fourth graders selected significantly more positive than negative cue instances on the transfer task, indicating they had indeed abstracted attributes from instances. Kindergarteners, however, selected precisely half of either cue instance, indicating failure to abstract the criterial attribute. Results contradict an assumption common to concept-shift studies. Experiment II was designed to show whether subjects attend to the dimensionality of a discriminated attribute in a concept-shift task. The subjects were 20 kindergarteners and 20 fourth graders assigned randomly by age group to an intra-dimensional or to an extra-dimensional shift, where dimensions were the same as in original learning, but all the cues were new. The subjects at both age levels performed significantly better on the intra-dimensional shift, supporting an attention-to-dimensionality hypothesis. Results were related to the concept-shift literature. (Author)