Road detection is a key task in developing an intelligent vehicle. Typical methods using machine vision suffer from being easy to be influences by outer conditions, so here we designed a new method to detect roads by a laser radar. Considering the intelligent vehicle may face much trouble in unsmoothed road, in road with anomalous edges or running crosswise with the road direction, we present a method named rectangle-searching algorithm to deal with these problems. In this algorithm, a rectangle is found out by the basis that the number of points involved in the rectangle should be as many as possible, while the size of the rectangle should keep small enough. After finding out the rectangle, an algorithm based on standard deviation is introduced to extract road edge points from road lines, which may help getting the road line from the rectangle. Experimental results validate efficiency and effectiveness of the algorithm.