Since the initial discovery of the non-exponential mass fractionation (non-EMF) of Nd isotopes analysis in 2002, similar deviations from an EMF pattern have been reported for measurements of a number of isotope systems (e.g., Si, Ge, Sr, Sn, Ba, Yb, W, Os, Hg and Pb) with MC-ICP-MS. However, the previous controversial reports on the magnitude of the deviations from EMF suggest that instrumental mass bias behaviour of MC-ICP-MS is neither fully understood nor well-characterised. Consequently, the standard approach of using a mass dependent fractionation (MDF) correction model (e.g., exponential law) may lead to both inaccurate and imprecise results. In this study, we systematically characterise the instrumental mass fractionation of MC-ICP-MS using Nd isotope measurements carried out under different plasma conditions, quantified using the normalised argon index (NAI) as an estimate of plasma temperature. Our results indicate that the mass bias of MC-ICP-MS is not always a simple exponential function of mass but shows systematic deviations from an EMF behaviour, which are closely associated with decreased NAIs. As a result, the conventional exponential correction yields a 143Nd/144Nd value of 0.512257 for the reference material BHVO-2 when the NAI is low, which is 722 ppm lower than the reported value of 0.512979. By tuning the plasma to higher NAIs (higher plasma temperatures), the deviations from the EMF array are systematically attenuated and the exponential correction is able to correct for the instrumental mass bias under high NAIs. In contrast, a regression correction model for Nd isotopes is developed to account for the observed mass fractionation behaviour that does not follow EMF under low NAIs, given that the regression correction relies on the observed loglinear fractionation of different isotope pairs and does not require both isotope ratios to undergo EMF. We expect that the analytical protocol and fundamental insights gained in this study are applicable to a wide range of other isotope measurements with MC-ICP-MS.