Real-time dim small targets detection in heavy cloud clutter is one of the key techniques of infrared search and track （IRST）system. An efficient and practical adaptive spatial-temporal filtering targets detection algorithm is proposed.At first a modified recursive adaptive temporal variance filter is employed to reject most background pixels and detect candidate targets which include few cloud edge pixels.After preprocessing an adaptive spatial filter proceeds to eliminate the residual clutter pixels according to their larger pixel-edgy intensity versus targets.This detection algorithm updates parameters as new frame arrived，thus is robust to scene variation.The usefulness of the algorithm is tested using five kinds of real-world IR image sequence with different background characteristics and proved to be effective and robust.