Since the beginning of the COVID-19 pandemic, adherence to epidemiological measures was proved related to the information exchanged through personal networks. However, there is still limited evidence on the effect of specific network properties. Using data from a randomised national sample of adults in Croatia (N = 765), we examine the role of various personal networks attributes with regard to three dependent variables: risk assessment of COVID-19, adherence to epidemiological measures and intention to avoid vaccination. We propose "pandemic discussion networks" as personal networks that encompass social contacts with whom respondents exchange pandemic-related information. Network heterogeneity in terms of the discussants' education level contributed to more protective behaviour, while network heterogeneity in terms of age contributed to an inclination towards vaccination. These associations were confirmed independently of behavioural homophily in terms of risky health behaviour, which was also found.