Grid-Connected Photovoltaic (GCPV) system is a type of photovoltaic (PV) systems which has been widely used as a renewable-based electricity generation. Nevertheless, the intermittency and fluctuation in weather conditions have caused inconsistent and varying output performance of a GCPV system. This paper presents a Multi-Layer Feedforward Neural Network (MLFNN) model for predicting the AC power from a GCPV system. Harmony Search (HS) was also employed to optimize several MLFNN parameters such that the prediction error could be minimized. The AC Watt-output of a GCPV system was predicted using MLFNN with solar irradiance, ambient temperature and operating PV module temperature as its inputs. These data were collected from a GCPV system located at Green Energy Research Centre (GERC), Universiti Teknologi MARA, Malaysia. In optimizing the MLFNN, HS was introduced to determine the optimal number of neurons in hidden layer, the learning rate and the momentum rate during training. After the training, testing process was conducted to validate the training process. In both training and testing, the prediction performance was quantified using Root Mean Square Error (RMSE). The performance of the HS-MLFNN was later compared with the performance of an Evolutionary Programming (EP)-MLFNN in predicting the AC power. The results showed that the hybrid HS-MLFNN had outperformed the hybrid EP-MLFNN by producing lower RMSE during both training and testing.