To suppress the mutual affects among different structure features of retinal and improve the detection precision of retinal microaneurysms, a microaneurysm extraction algorithm by fusing relationship among features was proposed. Firstly, the mean filter was applied to a retinal grayscale image, both the circular border and optic disc were detected, and the optic disc mask was created. Then, the green component of the retinal image was equalized with an adaptive histogram and Canny method was used to extract the edges before removing the image circular border and to fill the enclosed small area objects. Finally, with consideration of the relationship among different features, larger area objects were removed and an ‘AND’ logic was used to remove the retinal exudates, blood vessels as well as optic disc to obtain the retinal microaneurysm image. Experimental results indicate that the proposed method can effectively extract the microaneurysms in the retinal fundus image, and their sensitivity, specificity, positive predictive value and accuracy are 94.81%, 96.04%, 91.64 % and 95.66%, respectively. It can satisfy the clinical application requirements for strong stabilization and higher precision.