Abstract: In order to improve the efficiency of color-based image retrieval, this study proposed to apply multi-subblock strategy algorithm in image retrieval with sharp dominant colors. Multi-subblock strategy and subblock matching are helpful to control retrieval granularity and locate subject screens displaying the contents. On this basis, weight feedback of subblocks is added and repeated retrieval is conducted so as to capture the users' intents and improve the retrieval accuracy. Key issues of multi-subblock strategy, selection for color space, improvement for vector quantization and renewal of weight matrix was analyzed and a prototype of retrieval system was established while contrast experiments were launched. Experimental data prove that compared with Global Retrieval Method and Simple Subblock Cumulative Histogram Retrieval Method, such a method combining multi-subblock dominant colors and relevance feedback can improve retrieval precision rates.