We motivate and study the use of the forward kinematic model in the calibration scheme of parallel robots. Classical methods uses this model based only on the internal sensors for parameters elimination. One aim of the paper is to demonstrate how, in this well-constrained case, the model has numerous different solutions, how selecting a solution can lead to critical difficulties and possibly erroneous results. We propose two kind of alternatives, one drawing upon the methods of computer algebra to explicit the relation between the set of sensors and the parameters, and another which minimize the residual errors of an over-constrained formulation of the forward kinematics. This propositions are substantiated by experiments with planar robots, which are easy to describe completely and offer a full understanding of the underlying modeling and behavior.