Observation is the ‘gold standard’ for assessing parent-child behavior, however few studies have compared coding live, in real time, versus coding from videotapes in terms of their achievable levels of coder reliability within the field of parent programme research. This is important for practitioners and researchers for whom decisions might be influenced by time and financial constraints, but where outcomes may have real practical and clinical implications. Trained coders in the Dyadic Parent-Child Interaction Coding System Revised, coded 40 half-hour videotapes of 33 parent-toddler dyads interacting in the home on 29 items of dyadic behaviour. Four theorised composite variables were constructed. Videotaped data were compared to data drawn from the same interactions previously coded ‘live’ in the home. Correlations indicated significant agreement between the two modes at the item by item level (p < .001). Wilcoxon Rank tests revealed significant differences (p < .001) between the two modes. Eight items exceeded a ±30% change in median score suggesting clinically relevant differences. Although both methods achieved acceptable levels of inter-rater reliability, video coding achieved higher levels of agreement. Subtle differences exist between the two modes. Whilst neither mode proved superior it is suggested that they should not be used interchangeably.