Recently, several experiment-based methods have been developed for identifying rigid-body dynamic properties, such as the centre of gravity and moments of inertia. Among the newly developed methods, the inertia-restraint method is known to be relatively insensitive to measurement noises. However, the accuracy of this method depends heavily on the selection of response measurement points and excitation conditions, because their related matrices require several inversions. This study focuses on how to select measurement points and excitation point/directions to minimise the errors of identified inertia properties. To accomplish this, a statistical analysis of errors generated through the process of inertia property identification is performed, and the major factors influencing their accuracy are derived. An attempt is made to express the derived factors formed by the vectors of excitation and measurement conditions in physical quantities, such as length, area, and volume. From the investigations, the best and worst configurations formed by the selected measurement points are proposed, as well as some guides for the selection of excitations. Furthermore, a minimal number of measurement points and excitations are derived, and all suggested ideas are tested experimentally.