We consider the problem of fitting a conic to a set of 2D points. It is commonly agreed that minimizing geometrical error, i.e. the sum of squared distances between the points and the conic, is better than using an algebraic error measure. However, most existing methods rely on algebraic error measures. This is usually motivated by the fact that point-to-conic distances are difficult to compute and the belief that non-linear optimization of conics is computationally very expensive. In this paper, we describe a parameterization for the conic fitting problem that allows to circumvent the difficulty of computing point-to-conic distances, and we show how to perform the non-linear optimization process efficiently.