Brushless DC motors (BLDCMs) have been intensively used in automobiles, industry automations, and aeronautics and astronautics due to their advantages including high efficiency, high power density, and low noise. High resistance connection (HRC) fault is one of the typical motor faults. A severe HRC fault will cause serious temperature rise and even fire, and hence diagnosis of HRC fault in BLDCM is of significant. Generally, the HRC faults are detected by analyzing the motor current and voltage signals. However, there still deficiencies in the existed methods. This study designs a new method that combines the analysis of the array leakage flux signals and machine learning technology to realize location and quantitative analysis of HRC fault in BLDCM. First, multichannels of flux signals captured by a Hall sensor array that installed on the motor shell are sampled. A neural network model based on the timedomain features is designed to detect and localize the HRC faults. Subsequently, another neural network model based on the frequencydomain features is used to quantitatively analyze the HRC fault degree. The experimental results indicated that the accuracy of fault detection and localization is 9875% and the averaged root mean square error of quantitative analysis is 0018 Ω. The proposed method is noninvasive, easy to implement with high efficiency, hence, it will improve the accuracy and efficiency of HRC fault detection in BLDCM.