Classical models of infectious diseases such as the SIR model are instructive because they provide a 'mechanistic' modeling of the dynamics of the disease. It is, however, practically hard to estimate the involved parameters from a sample (i.e. from routinely collected surveillance data) as - among other factors - they usually do not include information on the number of susceptibles. This problem gives rise to another approach, where the underlying mechanism is disregarded, and the aim is simply to provide the best possible model of the time series of the number of new cases, including information on the time of the observation, describing seasonality and secular trends (parameter-driven models) and perhaps on past observations as well (observation-driven models). These are typically formulated within a regression framework, such as generalized linear models. I will introduce the foundations of such time series models, and illustrate them on real-life surveillance data. As a practical application of such models, I will touch the topic of prospective outbreak detection (which sometimes involves the mixing of the two approaches).