Synthetic aperture radar automatic target recognition (SAR ATR) methods fall short with limited training data. In this letter, we propose a causal interventional ATR method (CIATR) to formulate the problem of limited SAR data which helps us uncover the ever-elusive causalities among the key factors in ATR, and thus pursue the desired causal effect without changing the imaging conditions. A structural causal model (SCM) is comprised using causal inference to help understand how imaging conditions acts as a confounder introducing spurious correlation when SAR data is limited. This spurious correlation among SAR images and the predicted classes can be fundamentally tackled with the conventional backdoor adjustments. An effective implement of backdoor adjustments is proposed by firstly using data augmentation with spatial-frequency domain hybrid transformation to estimate the potential effect of varying imaging conditions on SAR images. Then, a feature discrimination approach with hybrid similarity measurement is introduced to measure and mitigate the structural and vector angle impacts of varying imaging conditions on the extracted features from SAR images. Thus, our CIATR can pursue the true causality between SAR images and the corresponding classes even with limited SAR data. Experiments and comparisons conducted on the moving and stationary target acquisition and recognition (MSTAR) and OpenSARship datasets have shown the effectiveness of our method with limited SAR data.