Image Retrieval in Data Stream is becoming very important in recent years with the development of internet and the focus on information security. In Data Stream, image retrieval is different from the traditional image retrieval in that, it requires high efficiency and network adaptive ability, while the traditional image retrieval does not have such hard requirements. In this paper, we propose a novel approach to retrieval image in data stream using Principle Component Analysis(PCA). By subdividing image into several blocks and extracting image principle features, our method can retrieval image very efficiently, this meets the needs of wide band width network traffic. Because the natural character of PCA, our method can be adaptive to the network traffic by simply modify a few parameters by an automatic way. Our experiments show that our approach has high Recall Rate and very low False Positive Rate.