Abstract: An automatic and precision method based on multi-features of plant was developed for weed detection. The color feature was used at first to segment green plant and soil background. Then the position feature was utilized to detect between-row weed and the texture feature was adopted to classify intra-row weed. Finally, the morphology feature was used to post-process the misclassified crop and weed. Images taken from the real wheat field (3～5 leaves seedling stage and within the different numbers of crop row) were used to test the novel solution of weed detection in the laboratory. The correct classification rate of crop and weed was over 89% and up to 98%. And the processing time was from 157 ms to 252 ms. Experimental results show that the weed detection method based on multi-features has a high classification rate and a quick processing speed.