A method for estimating geometric primitives of objects using 3D range data from a scanning laser is presented. It has been developed in order to position a robot arm relative to objects. An interactive segmentation technique selects a subset of 3D range data corresponding to an object and a parametrized model is fitted to the selected data using orthogonal distance regression. This yields a nonlinear least squares problem that is solved by the iterative damped Gauss-Newton method. The method is applied to the problem of estimating primitives corresponding to cylinders and experimental results are presented.