Abstarct: The purpose of the current research is automatic road sign detection and recognition. In first step we extract traffic sign from within various objects in the image. In the second step we recognize the type of the sign. In order to reduce illumination variation effect on detection step, we convert input image of RGB color space to HSI space. Then we extract red and blue regions of the input image by thresholding on hue and saturation parameters. Furthermore for better extraction of red and blue parts, we use fuzzy logic. In the next phase, we extract parts of the image that are triangle, circle, rectangle and octagon and send them to the recognition module. We produced a data-set of road signs that exist in Iran, containing 350 road signs in 35 different classes. Detection accuracy that we attained is 97.87 percent on this data-set. For exact classification of the sign, background of the detected sign is removed and then two approaches are examined. In the first approach, some features are extracted baxsed on profile of the sign image. Support vector machine were trained on 840 signs of 40 classes and 416 signs used for testing. Classification accuracy of 91.1 percent on test data was achieved. In the second approach, some features are extracted baxsed on gradient histogram and neural network is used for classification. To evaluate the proposed method, a data set of 1280 signs in 40 classes used. 67 percent of them, 864 signs assigned for training and 33 percent, 416 signs, used for test. Classification accuracy that we attained is 99 percent on train data and 93.86 on test data. Notice that our signs are in various light conditions, various backgrounds and various scales. Furthermore they are blur and have some rotation. Experimental results show the effective performance of the proposed method. In order to attain real time algorithm for detection and recognition of road signs, we converted early Matlab code to C++ code. We achieved a mean processing time of 170 ms per image in detection step and mean processing time of 121 ms per image in recognition step. These make the proposed method almost real-time.