A predictive EMC method for inductive coupling, using measurements carried out on a sphere encompassing a tested device, is at the moment under development in our laboratory. This method makes it possible to construct an equivalent source in the form of a sum of multipole terms. Once the equivalent sources of two devices have been designed, their inductive coupling can be calculated to prevent future EMC problems. Therefore, we have developed a measurement bench which allows the sensor to describe a spherical surface around the device we want to characterize. In this article, we present two methods for choosing the position of the measurement points (depending on the number of measurements and the chosen sensor) by minimizing the reconstruction error. The first one determines the position of the measurement points in order for them to be equidistant. In the second one, the source is considered to be random, and we maximize the determinant of the covariance matrix as a function of the sensor positions. This method allows us to define the best measurement points for a chosen number of points. This method is iterative and the number of measurements to be carried out is constrained by the maximum degree of spherical harmonics expansion. A comparison of the performances, advantages and disadvantages of these two methods is proposed and illustrated by an example.