Abstract: On the basis of different area density between tobacco stem and strip, a method for on-line detection of stem was developed by combining low-energy X-ray transmission imaging with real-time image analyzing and processing techniques, including morphological noise filter, gray degree threshold segmentation and attribution judgment. Optimal testing conditions were:the X-ray radiant intensity 70 ke V, the detection belt speed 1.6 m/sec and the segmentation threshold 40 000. The results of experimentally validation showed that the recognition rate for free stem with the diameter over 1.5 mm or lamina-attached stem was higher than 98%. When mixing the later into a flow of laminae, the recognition and rejection rates of stem reached 94.5% and 92.5%, respectively. The test results of threshed strips showed that for strips of grade Y1 in cigarette brand Furongwang, the rejection rate of stem reached 93.6% at a flow rate of 1 000 kg/hour, which indicated that on-line detection of the stem in strips by low-energy X-ray transmission imaging was feasible.