Neurofeedback (NF) training is a type of online biofeedback in which neural activity is measured and provided to the participant in real time to facilitate the top-down control of specific activation patterns. To improve the training efficiency, an investigation on the learning of EEG regulation and effect on neural activity during NF is critical. This article attempts to analyze the learning curve and the dynamics of the phase-locking value (PLV)-based brain network for a short-time EEG-based NF, in which 28 participants carried out alpha downregulating NF training in two consecutive days. The results reveal that participants could successfully construct the related learning network to achieve the training goals in the first day training and the beginning of the second day training. Moreover, the learning plateaus were discovered from the results of the relative amplitude and the functional brain network in the middle of the second day training. These findings could be helpful for better understanding of the learning process in NF from the functional connectivity viewpoint and would contribute to building a more efficient learning protocol for NF training.