A theoretical model designed to estimate the trait and method specific components of maternal and teacher ratings of childhood conduct disorder is presented. The model was applied to data collected on a birth cohort of New Zealand children. The model produced evidence to indicate that maternal and teacher ratings of childhood behaviour were contaminated by method variance. It was estimated that one-third of the variance in these scores reflected variance attributable to child behaviour traits. When the data were adjusted for the effects of method specific factors, the model suggested that conduct disorder measures were highly stable over time.