The dynamics of each link and joint of a robot is characterized by a set of 14 standard dynamic parameters (6 for the inertia matrix, 3 for the centre of mass coordinates, 1 for the mass and 4 for the drive chain inertia and friction). It is known that only a subset of the standard parameters, called the base parameters, are identifiable using the inverse dynamic model and the linear least squares techniques. Moreover, some of the base parameters are poorly identified because they poorly affect the joint torque. Thus they can be eliminated, leading to a new subset of dynamic parameters called the essential parameters. However, the identified values of the base or the essential parameters may be physically inconsistent regarding to the loss of the positive definiteness of the robot inertia matrix. Several methods have been developed in the past to verify the physical consistency of the identified parameters but they are complicated , time consuming and lead to non-optimal parameters. To overcome these drawbacks, a new method calculates a set of optimal standard dynamic parameters which are the closest to a priori consistent dynamic parameters obtained through CAD data given by the robot manufacturers. This is a straightforward method which is based on using the SVD and the Cholesky factorization and the linear least squares techniques. The new procedure is experimentally validated on an industrial 6 degrees of freedom Stäubli TX-40 robot.