Gear rattle noise outcome of drivetrains subject to input or output torque fluctuations are often evaluated experimentally after the product is designed and fabricated to allow measurements. This rattle noise avoidance using this method is not practical since any noisy system must be redesigned and fabricated for the next round of rattle evaluation. In this paper, an impact velocity-based rattle severity index is proposed to assess the rattle levels. Gear pair rattle experiments are performed to measure sound pressure levels of rattle noise as a function of excitation parameters (mean and alternating torque amplitudes, phases and frequencies) and gear mesh backlash amplitude. These sound pressure levels are compared to the corresponding rattle severity index values predicted by using a torsional drivetrain model. Predicted rattle severity index is shown to correlate well with the measured sound pressure levels within wide ranges of torque fluctuations and backlash magnitudes, allowing an assessment of rattle outcome of a drivetrain exclusively from a torsional model. Furthermore, the influence of types of vibro-impacts (single-sided, double-sided, periodic or non-periodic) on the resultant rattle noise is also shown to be captured by the proposed rattle severity index.