A crop row detection method based on machine vision was put forward to detect crop rows from farmland features images fast and effectively. In the process of image pre-processing, a new algorithm of the central line detection was proposed instead of the traditional vertical projection, and then a novel line detection method based on randomized algorithm was presented. Firstly, two different points are selected randomly from data space composed of locating points which determine a candidate line. Secondly, under the given distance tolerance, a strip region along the line direction is got, and the number of locating points in the region was accumulated. Lastly, the threshold rules are applied to determining whether the candidate line is the desired one. Test results show that the method can accurately find the crop rows under different light conditions and different growth stages. The image processing time is about 120 ms. Compared with Hough Transform (HT) and Randomized Hough Transform (RHT), the proposed algorithm has the advantages of smaller size of computer memory, shorter computational time and good robustness.