The quality and level of lint cotton are degraded because there are many foreign fibers and other harmful non-fiber trashes which are mixed into it in the process of plantation, production, transportation and machining. It will bring direct economic loss to textile industry. In order to improve the quality of lint cotton and increase the detection rate of foreign fibers, a pseudo-foreign fiber detection method based on visible spectrum machine vision was proposed. Lint cotton was made of thin layer after opening, and then transferred to the detection passage. Images of cotton layer with foreign fibers and pseudo-foreign fibers were snapshot by two line-scan cameras installed by the side of detection passage, and then it was stored into the industrial personal computers hard disk of experimental platform. Algorithms of image block and threshold were applied to extract pseudo-foreign fibers target areas, and statistical features in color, shape and texture of these target areas were calculated. Three classifiers: BP neural network, one to one directed acyclic graph linear kernel SVM and RBF kernel SVM were used to separate the two categories of cotton impurities. Results showed that 99.15% of the pseudoforeign fibers can be accurately identified, and the performance of RBF kernel SVM was the best among the three classifiers. With average recognition rate of 95.60%, the RBF kernel SVM can meet the online detection requirements of lint cotton trashes.