Nine 5-year-old infant school children, nine children with moderate learning difficulties, and nine children attending a school for the aphasic were selected on the basis of a counting test to judge relative number under three conditions: counting, guidelines, and control. The displays used consisted of two rows of dots and varied in numerosity and configuration. Length consistent displays were judged right by all children in all conditions on all numerosities. Many errors were made in judging length conflict displays. All groups judged better when they were told to count and when guidelines were added to the displays than in the control condition. Counting did not help more than guidelines. Small number displays were judged best in all conditions.