This study applies a back-propagation neural network to forecast fuzzy time series. Three models are proposed; a conventional fuzzy time series model and two hybrid models. Hybrid1 model uses a neural network approach to establish fuzzy relationships in fuzzy time series and hybrid2 model uses a neural network approach to improve forecasts from the conventional fuzzy time series model. The daily prices of golden pound for October 2014 were chosen as the forecasting target. The empirical results show that the hybrid2 model outperforms both the conventional fuzzy time series and the hybrid1 models.