Artificial Neural Networks have been successfully applied on time series forecasting problems in recent years. This paper aims to addrss the issues of applying Artificial Neural Networks on time series forecasting problems and introduce its current trend in the literature. We first divide the forecasting task into five sub-tasks including model selection, data transformation and editing, instance selection, input selection and network structures. Then the content of each sub-task is briefed, and the recent development of each sub-task is provided correspondingly. Future topics of each sub-task are also rendered.