Abstract: The primary objective of this research was to develop a rational, mechanistic constitutive model to predict the permanent deformation of subgrade pavement materials under dynamic triaxial repeated loads. A major advantage of the developed model is that it considers the permanent (plastic) as well as the resilient strain of the material which can be directly calculated from multi-layer elastic pavement response models. The model was developed based on static triaxial shear strength tests and dynamic repeated load triaxial test data conducted on four subgrade materials found in the State of Arizona. These materials included cohesive as well as cohesionless materials with the percentage of fines ranging from 1.2% to 31.5%. The materials investigated ranged from a plastic clay subgrade to a non-plastic sandy soil with a plasticity index from zero to 17.2%. In order to take into account the expected range of moisture variations in the field, all specimens were initially compacted at the target maximum dry density and optimum moisture content values according to the Standard Proctor method. A portion of these compacted specimens were tested directly after compaction; while, others were either soaked or dried and tested after a 24 hour conditioning period, in order for the moisture to reach an equilibrium condition. The developed model accurately predicts the permanent strain, as a function of resilient response, number of stress repetitions, degree of saturation, state of stress, material strength and a weighted plasticity index variable. The weighted plasticity index is defined as the material plasticity index multiplied by the percentage passing sieve number 200 (% fines). The model parameters were first generated for the cohesive materials and then for all materials investigated (cohesive and cohesionless). The model relates the &egr; _p/&egr;_r to the number of load repetitions by a power law. The slope of this power law is constant for the range of the materials investigated in this research while the intercept is a function of the stress to strength ratio, degree of saturation and basic material properties. The developed model was found to be rational, unbiased, accurate, and statistically sound. The model was based on a total of 3879 &egr;_p - N data points. The goodness of fit statistics of the final model selected was (logarithmic: Se = 0.143, S_e/S_y = 0.284 and R2_adj = 0.92 and arithmetic: Se = 2.454, S_e/S_y = 0.304, R2_adj = 0.91). An approximate methodology for subgrade (SG) rutting prediction, in pavement design practice, was also developed using the recommended study model. This method involves the use of the developed model and a multi-layer elastic program in order to predict the SG rutting. The predicted SG rutting values, using the developed method, was found to agree reasonably well with the SG rutting predicted using the Mechanistic-Empirical Pavement Design Guide (M-E PDG) on a series of Arizona Long Term Pavement Performance (LTPP) sections, used in this provisional feasibility effort. It was found that a correction (calibration) factor of 1.95 was necessary to be applied to the developed model rut depth predictions to have them agree with actual measured rut depth values. This calibration (correction) factor relates the developed model predicted SG rut depth to the M-E PDG predictions.