In the fault diagnosis of high-pressure common rail diesel engines, it is often necessary to face the problem of insufficient diagnostic training samples due to the high cost of obtaining fault samples or the difficulty of obtaining fault samples, resulting in the inability to diagnose the fault state. To solve the above problem, this paper proposes a small-sample fault diagnosis method for a high-pressure common rail system using a small-sample learning method based on data augmentation and a fault diagnosis method based on a GA_BP neural network. The data synthesis of the training set using Least Squares Generative Adversarial Networks (LSGANs) improves the quality and diversity of the synthesized data. The correct diagnosis rate can reach 100% for the small sample set, and the iteration speed increases by 109% compared with the original BP neural network by initializing the BP neural network with an improved genetic algorithm. The experimental results show that the present fault diagnosis method generates higher quality and more diverse synthetic data, as well as a higher correct rate and faster iteration speed for the fault diagnosis model when solving small sample fault diagnosis problems. Additionally, the overall fault diagnosis correct rate can reach 98.3%.