As data on the reliability and validity of ratings of infant temperament have accumulated, researchers have begun to ask what caregiver ratings really measure. An argument has been made that ratings of social behavior are less a reflection of enduring individual differences than a measure of rater characteristics and error variance. This study investigated the possibility that extraneous sources of variance would be found in Carey's Infant Temperament Questionnaire (ITQ). A large, diverse sample of mothers of 6-month-old infants completed ITQ's. Item analyses first identified a cluster of items left blank by a select, middle-class group of mothers. A large number of items (40 percent) were found to exhibit marked response biases, and parental characteristics were shown to affect response biases and styles, implying that systematic biases are built into the scale. Temperament ratings also varied with social class, ethnicity, and knowledge of infant development. Infants judged as being temperamentally easier tended to have mothers who were white, middle-class, and more knowledgeable than others. Results suggested that caregiver ratings on the ITQ reflect more than mere stylistic differences in infant behavior. (Author/RH)