Abstract: For the circular industrial CT scanned image slice of concrete that is dark in center and bright at periphery and the gray image without apparent bimodal feature, an image segmentation algorithm based on circular partition and region growth is designed to improve the accuracy of the image threshold segmentation. In order to reflect the superiority of the algorithm, comparative analysis with different segmentation methods is made. Then by applying MIMICS, after selective editing, hole-filling processing and redundant data removing, the Boolean operation is used to differentiate interface thickness of 50μm, so four-value division of the concrete CT slice is achieved. Furthermore, a finite element meso-level model is established by importing the four-value date into ABAQUS to simulate the Brazilian Splitting Test of concrete. The results show that the proposed segmentation method has similar results with the tests, so it can provide effective information for the construction of accurate concrete numerical model.