Research was conducted to determine the degree to which sampling vs. comparison processes account for age changes in referential communication. Children at three different grade levels communicated referents within related and unrelated word pairs. The word pairs used were from Cohen and Klein (1968). From their list of 30 related pairs, half were randomly sampled and served as the related pairs. The other half were used to compose unrelated pairs; to form these pairs, each similar nonreferent was replaced by a word dissimilar to the referent. The procedure consisted of asking children to communicate sinle word clues for each referent in the 30 word pairs. Clues with each word pair were then given to adult listeners who tried to identify the intended referent in each pair. Accuracy scores for each subject were based on the number of correct listener identifications. Twenty children in the 2nd, 4th, and 6th grades served as subjects. Half of the children at each grade level were randomly assigned to one of two experimental groups. The data from both groups indicated high accuracy at all ages on unrelated pairs and significant age improvement from second to sixth grade on related pairs. These results provide support for a comparison explanation of age differences in communication. Two additional studies were performed to evaluate two alternatives to the comparison explanation. Results from the three studies demonstrate younger children's communication difficulty in providing effective discriminative messages when required by the task. (Author/DB)