A deformable manipulator whose links can be bent according to different tasks is designed for unstructured environments in home service. The deformable manipulator can obtain relatively dexterous end-effector and extended workspace with much lower cost. However, frequent changes in link shape bring difficulties to the modeling and control of the manipulator. Firstly, the kinematic parameters of the deformable manipulator change significantly and irregularly. It is hard to measure the parameters directly due to the offset between the link coordinate and the link body. Secondly, the bending operations are frequent in order to deal with various tasks. Therefore, a less time-consuming calibration method is necessary, while the traditional calibration methods pursue high accuracy instead of efficiency. Finally, the calibration process for the deformable manipulator must be easy to be applied in home environment. To tackle above problems, a fast calibration algorithm for screw parameters of a deformable manipulator is proposed by using the circular motion of end-effector. When rotating any joint separately, the locus of end-effector is a circle in a plane, called the plane-of-rotation. The joint axis is perpendicular to the plane-of-rotation and goes through the centre of the circle. By using this property, the random sample consensus (RANSAC) and least-squares algorithms are used to fit the locus of end-effector and obtain the initial values of screw parameters. Then an extended Kalman filter (EKF) method is used to optimize the screw parameters with initial values and improve the calibration accuracy. The simulation and experimental results validate the effectiveness of the proposed method.