The dim and small infrared targets are easily covered for the background edge and strong noise, and the dim and small target detection is always the difficulties of infrared image processing. Based on analyzing the infrared image model, max-median filter algorithm is introduced to inhibit isolated noise point and improve signal to noise rate (SNR) under the condition of no influence on edge sharpness. And the dim and small target is detected at last using variance K-means clustering algorithm. Experimental results show that comparing with traditional filtering and morphologic algorithms, the algorithm has simple calculation, which can effectively get the dim and small target．