For the real-time detection of small target in the IRST systems, a temporal-spatial fusion filtering algorithm is proposed. The algorithm suppresses background and enhances targets by morphologic Tophat transform in spatial, and detects moving target by improved frame difference method in temporal. After the fusion image segmentation of the temporal and spatial processing result, the real target can be detected by the relation of the possible target＇s position in adjoining frames on the basis of the target＇s moving continuity and regulation. The algorithm entirely considers the characteristic that the small moving target can be detected in temporal and spatial, it enormously increases the target＇s SNR after the temporalspatial fusion enhancement. The detecting results of the true airplane＇s infrared image sequences in cloud background show that the algorithm can be more effective to detect low SNR moving small targets in complex background and reduce false alarm. The algorithm is easy to realize in hardware and can be applied effectively in the real-time target detection of the IRST systems.