The relationship between crash pulse shape and injury risk has been studied primarily with laboratory studies, but these are not necessarily representative of most real-life crashes. For the past decade, pulse information from real-life crashes has been available through event data recorders. The aim of this study is to evaluate how crash pulses from event data recorders can be parameterized with as few parameters as possible without losing the ability to accurately predict occupant injury. Pulses from 122 NASS/CDS cases with a delta velocity over 40 km/h were parameterized using eigenvector analysis. Six different pulses were created for each of these cases, including the original pulse and five approximations with gradually more pulse information. Using a finite-element sled model with the detailed THUMS human body model, the risk of chest injury was evaluated for each pulse version in each case. By comparing the results from each pulse approximation to the original pulse, the change in chest injury could be evaluated as a function of pulse approximation for each case. Using linear regression to analyse the chest injury error results it was found that a pulse with as few as four parameters-delta velocity, duration, and two shape parameters-can sufficiently describe the pulse shape from a chest injury point of view.