The assessment of infant temperament has been typically accomplished with parent questionnaires. When compared with temperament behaviours observed in the laboratory, parents and observers generally do not agree, leading some researchers to question the validity of parent report. This paper reports on a representative sample of infants whose families resided in non-metropolitan counties and whose temperament was measured in three ways: (1) standard parent report (Infant Behavior Questionnaire); (2) observer ratings across two lengthy home visits; and (3) observer coding of second-by-second reactions to specific emotion-eliciting tasks. In order to account for both trait and method variance, structural equation modelling was applied to a sample of 955 infants (M age = 7.3 months) using variables from the three methods that reflected the dimensions of positivity and negativity. Although models based solely on method factors and trait factors fit the data well, results indicated that a model that included method and trait factors provided the best fit. Results also indicated that parents and observers (either across the home visit or to specific tasks) converge, to a degree, on ratings of the positivity dimension but diverge on the negativity dimension. Copyright (C) 2008 John Wiley & Sons, Ltd
