The social environment of dairy cattle is important for their welfare under modern housing and management conditions. Social tension can negatively affect individuals even in a well-designed and healthy environment whereas affiliative behaviour may improve their well-being. The complex social relationships in a herd can be described comprehensively using network analysis. However, no up-to-date guidelines exist regarding the suitable time scale for assessing affiliative and agonistic behavioural structures in dairy cow groups. Dominance has been studied widely but the role of grooming and the relationships between affiliative and agonistic behaviours are still unclear. Furthermore, no measure exists that combines affiliative and agonistic interactions to describe the complete social experience of cows. In this study, we used video recordings and continuously assessed affiliative and agonistic interactions by all lactating Holstein cows in two groups (11 or 14 cows) during two periods (6 months apart). Based on the results of exploratory analysis for 5 days in one group, we aggregated the interactions over 3 days in each group and period, and analysed directed and weighted social networks for the group as well as individual level assessments of affiliative and agonistic behaviours. The affiliative and agonistic networks had no correlations, indicating that the two behaviours followed different patterns. Individuals in the agonistic networks were tightly connected with several reciprocal displacement relationships, whereas they were loosely connected in affiliative networks and the majority of these relationships were asymmetrical. Cows showed high between-individual variability in terms of their number of interactions and partners, indicating that specific cows had different roles in both social networks. We developed a specific balance index to describe the complete social experience of cows based on received and given affiliative and agonistic interactions. A comparison of the balance index with grooming and dominance indices found no linear associations, which implies that combining affiliative and agonistic behaviours may provide different insights to grooming or dominance alone. All three indices showed a moderate to strong stability over 6 months in the subgroups of cows present during both observation periods, indicating that the cows had stable social behavioural characteristics. Our results highlight the importance of directed networks when studying social relationships in dairy cattle groups. We suggest that the combined analysis of affiliative and agonistic behaviours using a measure such as the balance index might help to better understand how the welfare of individuals is related to the group in which they live.