A general model for describing the structure of multiple trait measures reported by multiple sources and observed at multiple times is developed. This model is applied to data on conduct disorder/oppositional behaviour, attention deficit and anxiety/withdrawal reported by three sources (child, mother, teacher) at three ages (8, 10 and 12 years) for a birth cohort of 1265 New Zealand children. This model was shown to fit the observed data adequately. The major conclusions of the analysis were: (a) reports of child behaviour derived from a specific source at a given time were of limited validity as measures of the child's behavioural tendencies; (b) reports of child behaviour derived from the same source were subject to correlated errors of measurement; (c) there was evidence of strong correlations (r = .80-.85) between conduct disorder/oppositional behaviour and attention deficit and weaker correlations (r = .30) between externalising traits and anxiety/withdrawal; and (d) when due allowance was made for errors of measurement in the report data, conduct disorder/oppositional behaviour, attention deficit and anxiety/withdrawal emerged as having very high stability across measurement periods. The implications of the results for a range of issues relating to the validity of report data, the interpretation of inter-informant correlations, the correlation/comorbidity of child behaviour problems and the stability of behaviour are discussed.