MambaMIL is a novel method for Multiple Instance Learning (MIL) in computational pathology that enhances long sequence modeling with sequence reordering. The method integrates the Mamba framework, which enables linear-time sequence modeling, into MIL to address challenges in long sequence modeling and overfitting. MambaMIL introduces the Sequence Reordering Mamba (SR-Mamba), which is aware of the order and distribution of instances and excels at capturing long-range dependencies among scattered positive instances. SR-Mamba is designed to learn correlations between instances in both sequential and transpositional orderings, significantly enhancing the original Mamba's ability to capture discriminative features. The method is evaluated on nine diverse datasets and demonstrates superior performance compared to state-of-the-art MIL methods in survival prediction and cancer subtyping tasks. The results show that MambaMIL achieves higher accuracy and robustness, with significant improvements over existing methods. The framework is effective in mitigating overfitting and computational overhead, making it suitable for computational pathology applications. The code is available at https://github.com/isyangshu/MambaMIL.MambaMIL is a novel method for Multiple Instance Learning (MIL) in computational pathology that enhances long sequence modeling with sequence reordering. The method integrates the Mamba framework, which enables linear-time sequence modeling, into MIL to address challenges in long sequence modeling and overfitting. MambaMIL introduces the Sequence Reordering Mamba (SR-Mamba), which is aware of the order and distribution of instances and excels at capturing long-range dependencies among scattered positive instances. SR-Mamba is designed to learn correlations between instances in both sequential and transpositional orderings, significantly enhancing the original Mamba's ability to capture discriminative features. The method is evaluated on nine diverse datasets and demonstrates superior performance compared to state-of-the-art MIL methods in survival prediction and cancer subtyping tasks. The results show that MambaMIL achieves higher accuracy and robustness, with significant improvements over existing methods. The framework is effective in mitigating overfitting and computational overhead, making it suitable for computational pathology applications. The code is available at https://github.com/isyangshu/MambaMIL.