It was wished to determine the relative reliabilities of graphic and semantic differential factors to evaluate applicability of two rival models to the measurement of pupil attitudes. Data was collected from a sample of sixth grade pupils on both abstract graphic and typical semantic differential scales. Reliability estimates were constructed from achieved communalities and multiple correlations. Mean reliabilities were compared among factors and pupils' reading skill level groups and it was concluded that semantic differential scales produced reliabilities higher than those derived from graphic differential scales. It was decided to drop further development of the abstract graphic differential. (Author)