Infrared small target detection (IRSTD) has focused on improving U-Net skip connections, but challenges like target information loss and poor detection of target contours remain due to low contrast between infrared small targets (IRST) and the background. We propose a Bidirectional Feature Fusion U-Net (BDFU-Net) with a Bidirectional Fusion Block (BDF) in the skip connections to address these issues. The BDF integrates features using attention mechanisms to enhance discriminability and applies Down Feature Fusion (DFF) to filter background noise and highlight target features. Up Feature Fusion (UFF) merges deep and shallow layers to restore target details and locations. Experiments show our method outperforms existing approaches on three public datasets.