RF fingerprinting (RFF) is a non-encrypted authentication technique that provides an additional layer of security for wireless devices, which has a promising application. However, existing RFF recognition techniques that rely on deep learning (DL) are usually available with limited equipment. In actual application scenarios, new wireless devices are constantly appearing, such as unknown drones that appear suddenly in the sky. In such cases, the RF monitoring system should equip the ability to discover the unknown device (i.e., open-set recognition (OSR)) and use captured few samples of new devices incrementally updating knowledge of the system. This requirement brings two challenges: 1) incremental updates from few-shot samples are prone to lead to catastrophic forgetting and over-fitting problems; 2) constructing a reliable OSR mechanism for new devices with few-shot samples is difficult. To tackle this challenge, for the first time, we propose a novel few-shot open-set incremental learning (FSOSIL) framework via meta-learning for RFF recognition (Meta-RFF ). The core idea of Meta-RFF is to simulate few-shot RF signal incremental learning by constructing many pseudo-FSOSIL tasks. In particular, to strengthen the OSR capability, we further design RF feature augmentation and open space learning modules. The algorithm is validated on the large-scale aircraft recognition dataset (namely ADS-B), which shows that the close-set accuracy and open-set AUROC of the new class improve the performance by about 10–20% compared to other algorithms with 1-shot. And in 10 increments, our algorithm possesses a lower performance decay rate (about 3%).