A novel method for detecting obstacles in grass is proposed by fusing information from ladar and camera. Firstly, training range data is acquired by a 3-D ladar, and shape feature parameters of obstacles and grass are calculated respectively. Using Expectation Maximization (EM) algorithm, the Gaussian Mixture Model (GMM) which represents the distribution of shape features is learned. With these learned models, the 3-D points of candidate obstacles in the scene can be obtained. Meanwhile, mean-shift algorithm is applied to the corresponding color image to acquire the region information of the scene. With the result of joint calibration of the 3D ladar and the camera, those candidate obstacle points are projected to the segmented image and fusion is then performed to make the final decision. Experimental results show that the proposed method can distinguish obstacles from grass effectively with high precision and low false-alarm.