Abstract: In processing multimodal magnetic resonance imaging (MRI) brain tumor images, existing segmentation models with dual-encoder structures often have difficulty adequately exploiting the features of each modality, whereas a four-encoder structure tends to cause a noticeable increase in parameters and computational cost. To address these issues, this paper proposes an asymmetric multi-encoder algorithm for multimodal brain tumor segmentation (AMM-Net). The proposed model fully extracts features from different modality combinations through three encoders, and utilizes two fusion strategies in the encoding stage to comprehensively integrate the features extracted from the three encoders. Finally, an attention gating mechanism is introduced during the decoding process to retain more detailed information. Experimental results on the BraTS2018 dataset show that the segmentation accuracy(Dice score) of this model reaches 0.795 for the enhancing tumor (ET), 0.891 for the whole tumor (WT), and 0.828 for the tumor core (TC). Moreover, the segmentation performance of this model in the enhancing tumor region is significantly superior to that of current mainstream segmentation models, and it also achieves favorable segmentation results in both the whole tumor and tumor core regions