Effective receptive field is crucial for accurately detecting remote sensing objects. However, the universally employed Convolutional Neural Networks suffer from a constrained receptive field, while the substantial computational demands of Transformers render them impractical for processing high-resolution remote sensing imagery. To this end, we propose Cross-Scan Mamba (CSM) based on Mamba, which is a selective state space model (SSM) capable of dynamically adjusting to input in order to efficiently model long-range dependencies. To address the issue of uneven and discontinuous target distribution in remote sensing images, we propose a Cross-Scan Mamba Block (CSMB): a framework that effectively integrates continuous scanning, regional scanning, and skip scanning methods to model the rich global context and local features within remote sensing scenes. Compared to a single scanning method, CSMB combines different scanning approaches to reduce the required scanning distance, while also introducing large convolutional kernels to supplement spatial features, enabling rapid and precise recognition of remote sensing targets. CSM showcases state-of-the-art performance on benchmark datasets, notably achieving an impressive 81.07% mAP on DOTA-v1.0 and 90.68% mAP on HRSC2016, thereby substantiating its efficacy in remote sensing object detection endeavors.