A 29-item sociometric nominations instrument (Guess Who) was administered to 13,045 children in 603 elementary classes. An image analysis of the frequencies of nomination on the items yielded four factors: Disruptive, Bright, Dull, and Quiet-Well Behaved. Four scaling methods were compared with regard to class-size bias: raw frequencies, division by class N, binary truncation, and standardization within classes. Binary item-frequency truncation yielded scores which were demonstrated to have superior external validity using diagnostic groups, self-reports, and teacher ratings as criteria. Internal consistency reliabilities for the four factor scales ranged from .61 to .77, while split class reliabilities ranged from .56 to .74.