Seasonal forecasts aim to provide physically-based probabilistic outlooks of the climate conditions for the coming seasons. We create them by initialising coupled atmosphere-ocean-sea-ice models with observations and integrating ensembles of them forward in time. As such they can be considered extensions of numerical weather predictions. Slowly varying surface temperatures over the oceans, especially in the tropics (e.g., ENSO) provide important sources of atmospheric predictability on seasonal timescales. The resulting predictions can provide useful early warning of anomalous climate conditions but also bear difficulties with regards to model errors, small forecast signals and limited sample sizes. In this talk I will give an overview of seasonal forecasts and the opportunities and challenges they bring.