This study employed a multiple regression to predict examinees' differential performance when tested by familiar and unfamiliar examiners. Subjects were 32 preschool and school-age handicapped children, each of whom had been tested on the Clinical Evaluation of Language Functions, once by a familiar and once by an unfamiliar tester, within a crossover design. Teachers' ratings, subjects' self-reports, an anxiety scale, and a sociometric procedure accounted for nearly 40 percent of the variance in differential test performance. Implications for valid assessment of handicapped children are discussed. (Author)