One of the challenges to the implementation of the mechanistic-empirical pavement design guide (MEPDG) comes from calibrating the transfer functions. This paper focuses on calibration of one of the major distress models in flexible pavement: permanent deformation or rutting. Two key aspects are critical to a successful rutting model calibration: data and method. Regarding the data, existing in-field information only provides total rut depth, which could not meet the requirement of permanent deformation in each structural layer by the MEPDG. Concerning the method, existing work either fails to address calibration factors from a holistic perspective by only focusing on individual sections separately or ignores variability inherent in those factors. In this study, layer-wise permanent deformation from instrumented pavement under accelerated pavement testing serves to accommodate the models calibration. A systematic calibration procedure is established, which globally optimizes all available information across all test sections. Through simulation and numerical optimization, optimal calibration shift factors for three typical flexible pavement materials, asphalt mixture, unbound granular base, and finegrain soil are obtained as 0.60, 0.49, and 0.84, respectively. This implies that the uncalibrated MEPDG is biased toward overprediction of rut depth. It is further suggested that a more rational result for each calibrated factor is to introduce an appropriate distribution to characterize its uncaptured variability. In addition, a case study involving using calibrated MEPDG to predict pavement performance or life indicates that (1) model calibration has a significant impact on the prediction and (2) the "fourth power law" is supported by the MEPDG.