In the actual work of rotating machinery, it is difficult to achieve accurate fault diagnosis because of the limited fault samples. To address this problem, a small-sample fault diagnosis method based on GADF and PAM-Resnet is proposed. Firstly, the proposed method constructs a data enhancement strategy, which converts a small number of 1D signal samples into 2D GADF images, and then crops the GADF images into multiple sub-images to obtain a large number of image samples, which solves the problem of insufficient number of samples. Then, a position attention model(PAM) is constructed, which uses horizontal and vertical convolution to give weights to horizontal features and vertical features, respectively, and fuses the two features to obtain the position information of the GADF image. Finally, the PAM is inserted into the residual block to construct the PAM residual block, and multiple PAM residual blocks are used to construct the PAM-Resnet. The PAM-Resnet can effectively focus on location information and has a strong fault feature learning capability. The fault diagnosis experiments of gearbox and rolling bearing under the small sample environment are carried out respectively, and the results indicate that the proposed method has higher fault diagnosis accuracy and can accurately diagnose the fault types under small sample environment.