AbstractPredicting the peak ground velocity of vibrations is essential to blast mining operations in order to design the charge weights so as not to exceed certain thresholds that prevent damage to buildings and other infrastructure. The problem is usually marred by a large scatter of observed peak ground velocity due to unknown complexities of seismic wave propagation. Classic peak ground velocity prediction methods employ empirical formulas, the most widespread being the scaled distance approach that has the least parameters to calibrate and works for a single sensor. In this study, we used data from 55 mining production blasts recorded by an array of 81 seismic sensors in an open pit iron ore mine in Austria. We evaluated and compared different methods for predicting peak ground velocity. The large data set provides sufficient constraint to independently resolve the charge weight exponent c, the radial decay constant b and local site factors. The c/b‐ratio of 0.2 that we find for our site is far smaller than that implied by the US Bureau of Mining scaled distance method, and peak ground velocity predictions made with the latter approach are significantly worse. This highlights the importance of using site‐specific data to calibrate predictive models and suggests that relying on arbitrary priors may lead to inaccurate predictions. For the charge weight exponent, we find a value of 0.5 which we interpret as expression of the physical relationship among charge weight, energy and amplitude, suggesting that this may be a global, site‐independent, value. This result has probable a broader relevance beyond our specific location and could improve prediction outcomes on other sites.