The root is an important plant organ for nutrient and water uptake from the surrounding soil. To solve the technical problem of 3-D observation for plant roots in situ, sequence images of plant roots in situ were acquired at first with using X-ray CT scanner, and then the 3-D visualization of roots in situ was implemented in computer using image and graph processing methods. In order to improve the accuracy of segmenting root CT images, a compositive segmentation algorithm was proposed and implemented based on the spatial geometrical and morphological features of root systems. The algorithm could clean markedly up the pixels of non-root regions of which the gray levels were very similar to roots regions and the positions were very closed to roots regions. 3-D surface models of plant roots were reconstructed with marching cubes (MC) algorithm, and 3-D visualization of plant roots in situ was implemented with the tool box of VTK. It was proved by programming experiment that the approach and methods proposed were feasible and effective for 3-D visualization of plant roots in situ.