Cattle are considered to be the main reservoir for Vero cytotoxin-producing Escherichia coli (VTEC) O157, a cause of food-poisoning (and even death) in humans. Here, the transmission of E. coli O157 within a typical UK dairy herd is modelled using a semi-stochastic network model. The model incorporates demographic as well as infection processes. Indirect transmission is modelled homogeneously, while direct transmission is modelled via a dynamic contact network. The aim was to investigate the effects of heterogeneity and clustering on the prevalence of infection within the herd and discover whether, particularly in terms of choosing an intervention strategy, it is necessary to include heterogeneity in direct contacts when modelling this sort of system. Results show that heterogeneity in direct contacts can make it more difficult for the pathogen to persist, particularly when the average number of contacts (per animal) in each group is small. They also show that the relationship between clustering and prevalence is not simple. For example, increasing the average number of contacts can increase clustering and prevalence. However, when the average number of contacts in each group is sufficiently high, higher clustering leads to lower prevalence. It would seem that clustering can aid the flow of infection under certain circumstances, but hinder it under others (probably by preventing wider dissemination). Further results show that indirect transmission (as it is modelled here) effectively removes the effect of heterogeneity in direct contacts. In terms of investigating proposed interventions, the results suggest that a network model would only be required if there was evidence to suggest that direct transmission was the major source of infection.