Calibration aims at identifying the model parameters of a robot through experimental measures. In this paper, necessary mathematical conditions for calibration are developed, considering the desired accuracy, the sensor inaccuracy of the joint coordinates, and the measurement noise. They enable to define a physically meaningful stop criterion for the identification algorithm and a numerical bound for the observability index O3 , the minimum singular value of the observability matrix. With this bound, observability problems can be safely detected during calibration. Those conditions for calibration are illustrated through a simple example.