An age-dependent, serious-to-fatal (AIS3+), thoracic risk curve was derived and evaluated for frontal impacts. The study consisted of four parts. In Part 1, two datasets of post mortem human subjects (PMHS) were generated for statistical and sensitivity analyses. In Part 2, logistic regression analyses were conducted. For each dataset, two statistical methods were applied:​ (1) a conventional maximum likelihood method, and (2) a modified maximum likelihood method. Therefore, four statistical models were derived — one for each dataset/statistical method combination. For all of the resulting statistical models (risk curves), the linear combination of maximum normalized sternum deflection and age of the PMHS was identified as a feasible predictor of AIS3+ thoracic injury probability. In Part 3, the PMHS-based risk curves were transformed into test-dummy-based risk curves. In Part 4, validation studies were conducted for each risk curve. We used a theoretical field model — an array of modeled real-world crashes whose attendant risks were estimated by the subject thoracic risk curve — to compare the predicted injury rates with the actual injury rates from the field. The field data consisted of belt-only (no airbag) adult drivers in 11-1 o’clock, full-engagement, non-rollover, frontal, tow-away crashes from the National Automotive Sampling System. The assessments were conducted from two perspectives:​ (1) aggregate estimates of AIS3+ thoracic injury rates across the speed range for each age group, and (2) point estimates of thoracic injury rates for lower- and higher-speed crashes. Analysis of the field model predictions indicated that the statistical method selection was more important than the dataset selection. The two risk curves based on the modified maximum likelihood method demonstrated acceptable fidelity. Of those two risk curves, the one with the best fit for both the aggregate and the point-estimate predictions was chosen as the provisional risk curve.