The invention discloses a bearing fault diagnosis method and device under a sample imbalance condition, and relates to the field of bearing fault diagnosis, and the method comprises the steps: collecting time domain vibration signals of a bearing fault through equipment, and classifying the collected signals; segmenting the fault signals into a plurality of samples, then performing fast Fourier transform to obtain frequency domain data, and making a training set and a test set according to a proportion; building a VAE-GAN fault sample generation model, respectively inputting minority classes of fault samples in the training set into the model, and balancing the training data set; building an FLCNN fault classification model, and inputting the balance training set obtained in the step S3 into the model for training; and analyzing an experiment result. According to the bearing fault diagnosis method and device, the VAE network and the GAN network are combined, meanwhile, the feature coding capacity of the VAE for training data and the adversarial learning mechanism of the GAN are used for reference, and compared with other methods, the fault diagnosis precision can be effectively improved under different unbalanced proportions.