This work presents a method to estimate 3D shape of an active needle inside tissue using 2D transverse ultrasound images. The shape of the needle provides a valuable feedback information for precise control and guidance of the needle inside tissue toward target. We used a series of image processing techniques to identify the needle’s cross section in the ultrasound images. Using this method, we estimated the 3D shape of a tendon-driven active needle, when bent inside a transparent phantom tissue using a robotic needle insertion system. The estimated shape of the needle was then compared with true shape of the needle captured by two cameras. At least three ultrasound images were required to estimate the needle shape with a second order polynomial function. We found an average error of 0.54mm and a maximum error of 1.00mm in shape estimation compared to the true shape of the needle captured by the cameras for an insertion depth of 70mm.