The quality of parent-child interaction has consistently been associated with child psychosocial outcomes. Postnatal depression (PND) and anxiety (PNA) can have an adverse effect on the parent-child interaction. The Coding Interactive Behaviour (CIB) measure has frequently been used to assess parent-infant interaction quality; however, few have investigated the construct validity. The purpose of this study is to: (1) compare two theoretical models and one data-driven model for best fit and to test the best fitting one for measurement invariance among mothers with and without PND, and (2) compare effects of PND and PNA on the mother-infant interactions. The sample consisted of mothers with PND (n = 236) and a subclinical sample of mothers with elevated symptoms of PND but did not meet criteria for a diagnosis (n = 183) and their infants (aged ≤ 6 months). PND was assessed with a diagnostic interview and PNA using a self-report measure. Confirmatory factor analyses showed that the data-driven model, consisting of the composites ‘Maternal Sensitivity’, ‘Child Engagement’, ‘Maternal Social Withdrawal’, and ‘Maternal Controlling Behaviour’, reached the best fit and showed measurement invariance with regard to PND. Factorial ANOVAs showed no effects of PND and PNA on the composites. The study is the first to investigate and find measurement invariance in the CIB, providing evidence of construct validity. The importance of investigating the factor structure of an instrument liable to change due to contextual factors is discussed.