The Point Distribution Model (PDM) is useful for many tasks involving the location or tracking of deformable objects. However, non-linear variation must be approximated by a combination of linear variations, resulting in a non-optimal model which can produce implausible object shapes. The Polynomial Regression PDM improves on this by allowing polynomial deformation, but at the cost of computational complexity, and it still fails for objects in which bending or pivoting occurs. We propose an extension to the PDM which makes selective use of polar coordinates, and give examples to show that the models produced are often more compact and precise than either of the above methods. We also present two different algorithms for automatically determining pivot positions, and test them on both real and synthetic data.