Long-term time series forecasting is crucial in various domains, including weather, traffic, and energy. In a time series, the time domain contains intuitive time-varying characteristics, affording valuable insights into predicting trends and details, while the frequency domain harbors underlying periodic patterns, identifying pivotal elements within historical data essential for forecasting. However, prior methodologies often fixated exclusively on either the temporal or frequency domain, falling short in capturing the information within historical data in a holistic view for forecasting. To address the issue mentioned above, we propose a Temporal-Frequency Interactive Attention Network (T-FIA) to refine the modeling of time series by concurrently harnessing insights from both the temporal and frequency domains. In T-FIA, we formulate a temporal-frequency interactive encoder to capture domain-specific information from the dual perspectives of temporal and frequency domains. Subsequently, we facilitate information interaction between them, enhancing a more comprehensive understanding of time series and enabling precise time series forecasting. Experimental evaluations on seven datasets demonstrate the effectiveness of our proposed T-FIA.