This paper presents a newly designed scheme based on neural networks to detect loss of excitation (LOE) in synchronous generators. The proposed scheme uses more accurate mechanism and needs fewer parameters in order to achieve fast and reliable detection of LOE. Furthermore, being able to discriminate between LOE and stable power swings is a major concern to enhance the performance of traditional LOE protection. Therefore, the designed network is trained to discriminate between both cases clearly. For training and testing the proposed neural network, MATLAB program has been used for simulation. In addition, by using comparison analysis between the designed network and the previous ones and the traditional MHO relay, the results ensure that the proposed scheme has more secure and fast characters in detecting and discriminating LOE. Keywords Synchronous generator • Loss of excitation • Power swing • Protection • Neural networks • Artificial neural network • Stable power swings List of Symbols X s The system impedance X d The synchronous reactance of the d-axis X ′ d The transient reactance of the d-axis X ′′ d The sub-transient reactance of the d-axis X ′ q The transient reactance of the q-axis X ′′ q The sub-transient reactance of the q-axis T do ′ The d-axis transient open-circuit time constant T do ′′ The d-axis sub-transient open-circuit time constant T qo ′ The q-axis transient open-circuit time constant I Generator's current V Generator's voltage Abbreviations AI Artificial intelligence ANFIS Adaptive neuro-fuzzy inference system ANN Artificial neural network FACTS Flexible AC transmission systems LOE Loss of excitation