A Markov-modulated framework is presented to incorporate correlated inter-event times into the stochastic susceptible-infectious-recovered (SIR) epidemic model for a closed finite community. The resulting process allows us to deal with non-exponential distributional assumptions on the contact process between the compartment of infectives and the compartment of susceptible individuals, and the recovery process of infected individuals, but keeping the dimensionality of the underlying Markov chain model tractable. The variability between SIR-models with distinct level of correlation is discussed in terms of extinction times, the final size of the epidemic, and the basic reproduction number, which is defined here as a random variable rather than an expected value.1. Introduction. The SIR-model is a well-studied epidemic model that, together with its generalizations, has been widely applied to infectious diseases such as measles, chickenpox, or mumps, among other situations where infection confers (typically lifelong) immunity; a good discussion about SIR-models analyzed from deterministic and stochastic perspectives can be found in the lucid texts by Allen [1, 2], Andersson and Britton [8], Bailey [16] and Keeling and Rohani [36]. The SIR-model is first analyzed in depth by Kermack and McKendrick [37] in 1927 in order to study the evolution of a disease in a closed community of finite size where, at time t, individuals are classified into three categories: S(t) susceptibles, I(t) infectives, and R(t) removed individuals. The general description in [37] is related to an homogeneously mixed population (i.e., any infective can infect any susceptible with equal probability), where the infection and recovery rates of a given infective depend on the total time that this individual has been infected for. The analytical