To detect and localize the crop rows quickly and effectively for navigation of agricultural machines, a new algorithm for crop rows detection is proposed in this paper. A navigation software was developed in VC++ 6.0. Crop rows were separated from soil background by image pre-processing, and the localization points were got by vertical projection. The world coordinates of each localization point were computed according to the principle of perspective transformation and the camera calibration results. With the parallel characteristic of crop rows, an improved algorithm based on Hough Transform was employed for the detection and localization of crop rows. The experiment with images of crop rows and the simulation experiment in laboratory showed that the new algorithm took 219.4 ms to process a 320×240 pixels color image, and the average errors of navigation distance and navigation angle were 2.33 mm and 0.3°. The experimental results confirmed that the algorithm was accurate, effective and fast enough to detect and localize crop rows for real-time navigation.