At the core of a mechanistic-empirical (M-E) pavement design method is a collection of performance models that each predicts the development of a specific pavement distress, such as fatigue cracking and surface rutting. Each model has both mechanistic and empirical parts. The empirical parts need to be calibrated to remove bias and increase prediction accuracy. This process has traditionally been conducted with small numbers of field sections for which materials may or may not have been sampled and tested. This paper presents a new calibration approach that uses network-level field data and statewide distributions of material properties, without having to sample and test every individual calibration section. The calibration of the fatigue and reflection cracking models for CalME, the M-E design software developed for the California Department of Transportation, is used as an example to illustrate the new approach. The new approach works by correlating the statistical distributions of M-E design inputs with the statistical distribution of pavement performance, both at the network level. The uncertainties affecting pavement performance are divided into those specific to a given project (within-project variability) and those that vary between projects (between-project variability). This distinction allows a clear definition of design reliability. The results showed that the new approach can overcome some of the network-level data limitations and provides a reasonable calibration ready for routine pavement design.

Key words: infrastructure, pavements, design and rehabilitation of asphalt pavements, asphalt, mechanistic-empirical pavement design