Many pathogens, especially those responsible for emerging infectious diseases, are transmitted in a host community. How the host community structure affects an epidemic is still debated, particularly whether increasing the host community complexity would tend to amplify or dilute the incidence of an epidemic in a target population, e.g. humans or cattle. In this paper, we build a stochastic Susceptible–Infectious–Recovered model (SIR) and compare epidemiological dynamics in a target population between three simple host community structures with an increasing complexity. Globally, our results show two possible main outcomes. First, an intermediate host can have a diluting effect by preventing the direct transmission from hosts to the target population, thus reducing the prevalence of infection. Second, when the infection comes from two different sources (e.g. two different populations), the effects of the epidemic in the target population are generally amplified. By highlighting that the structure of the ecological hosts network can dramatically affect epidemics, our results may have implications for the control of emerging infectious diseases.