Transformer-based sequential recommendation models perform well, but the quadratic computational complexity of the attention mechanism limits their ability to model long behavior sequences. Recently, the Mamba architecture, with a selective state space model at its core, has been applied to sequence recommendation tasks to address the effectiveness-efficiency dilemma. However, it still has some issues. On one hand, due to the unidirectionality of the state space model, it cannot capture comprehensive item-item dependencies. On the other hand, it focuses solely on discrete time-domain patterns and ignores frequency-domain information with periodicity. To address these issues, we propose a time-frequency adaptive fusion modeling framework for sequence recommendation to effectively capture user behavior features. Experiments conducted on three benchmark datasets show that our model balances recommendation performance and efficiency, outperforming state-of-the-art baselines. Our code is available at https://github.com/liminss/TFAF4SR .