Sequential recommendation techniques aim at providing personalized recommendations by analyzing the dynamic preferences and sequential dependencies in user behavior sequences. However, dense datasets consisting of long sequences have the difficulty of fully capturing the sequential dependencies within historical user behaviors, while sparsity datasets of short sequences often struggle to extract sufficient interaction information due to the limited historical user behaviors. These challenges lead to undesirable recommendation performance and exacerbate the difficulty of achieving effective recommendations in sparse data scenarios. Then, we propose a novel sequential recommendation model named MaTrRec that integrates Mamba and Transformer. This model leverages the linear complexity advantages of Mamba while effectively enhancing sequential dependency modeling through its state-space representation and efficient sequential processing capabilities in continuous space, enabling the extraction of deeper sequential information from historical user behaviors. Moreover, it utilizes the global attention mechanism of Transformer to thoroughly explore the dependencies among items in historical user behaviors, thereby enhancing the model’s predictive performance and robustness across both long and short sequence datasets by effectively capturing both local and global contextual information. To further optimize the computational complexity of the model, we design a linear sequential recommendation model called MaTrRec*, which reduces the computational complexity through a linear attention mechanism and effectively balances the complexity and recommendation accuracy. Experimental valuations on five widely used public datasets demonstrate that MaTrRec and MaTrRec* outperform the state-of-the-art sequential recommendation models. Our MaTrRec can significantly achieve up to 8.49% higher model accuracy. Code: https://github.com/Unintelligentmumu/MaTrRec .