Understanding how individuals modify their social interactions in response to infectious disease is of central importance for our comprehension of how disease dynamics operate in real-world populations. Whilst a significant amount of theoretical work has modelled disease transmission using network models, we have comparatively little understanding of how infectious disease impacts on the social behaviour of individuals and how these effects scale up to the level of the population. We experimentally manipulated the parasite load of female guppies (Poecilia reticulata) and introduced fish either infected with the ectoparasites Gyrodactylus spp. (experimental) or uninfected (control) into replicated semi-natural populations of eight size-matched female guppies. We quantified the behaviour and social associations of both the introduced fish and the population fish. We found that infected experimental fish spent less time associating with the population fish than the uninfected control fish. Using information on which fish initiated shoal fission (splitting) events, our results demonstrate that the population fish actively avoided infected experimental fish. We also found that the presence of an infected individual resulted in a continued decline in social network clustering up to at least 24 h after the introduction of the infected fish, whereas in the control treatment, the clustering coefficient showed an increase at this time point. These results demonstrate that the presence of a disease has implications for both the social associations of infected individuals and for the social network structure of the population, which we predict will have consequences for infectious disease transmission.