Due to the limited availability of labelled data in many real-world scenarios, we have to resort to data from other domains to improve models’ performance, which prompts the advancement of research regarding the cross-domain few-shot image classification task. In this paper, we systematically review existing cross-domain few-shot image classification algorithms published in recent years. We categorize these algorithms into data-augmentation and feature-alignment paradigms and present their recent progress. We summarize three commonly-used cross-domain datasets for benchmarking few-shot image classification tasks and relevant scenarios. Finally, we outline existing limitations and future perspectives.