Positive versus negative instances, number of binary dimensions (three vs. five), and the number of values changed per trial (one vs. two) were compared in single-attribute concept attainment. College students judged the relevance of each cue on each trial and received only sufficient information for solution. Initially, positive instance performance was superior, but it fell below negative instance performance by the final trial. Positive instance errors arose more often from memory deficits and from choice of irrelevant dimensions. Negative instance errors arose when interpreting information about present stimuli and the relevant dimension. Results were interpreted through a logical analysis of the task.