Direct workplace whole-body vibration exposure assessment provides ecological validity for evaluating health risk in epidemiological studies, yet it is complex and expensive in practical applications. Exposure prediction modeling could be a cost-efficient alternative to directly assessing occupational vibration exposures. The objective of this study was to model directly measured whole-body vibration exposures with predictors from machinery, farm, and self-reported characteristics among Canadian prairies farmers. As per ISO 2631-1, whole-body vibration data were measured on the seat surface at three axes (x, y, z), then summarized into vector sums of the root-mean-squared (RMS) acceleration and the vibration dose value (VDV). All candidate predictors were obtained via questionnaires and onsite observations. A total of 87 whole-body vibration measurements were collected from 40 male farm workers located at 21 central Saskatchewan farms. Using log-transformed RMS and time-standardized VDV outcomes, modeling started from the bivariate analysis where predictors with P-values < 0.2 were considered eligible for multivariate analysis. With random effects of 'farm' and 'farmer', a series of mixed-effects models were constructed through the manual backward elimination method. Final models were internally validated by 1000 bootstrapped samples. The RMS model explained 47.7% of the variance in the directly measured RMS vector sum, with 42.7% obtained from five predictors of 'horsepower', 'transmission', 'vehicle year', 'jerk/jolt frequency', and 'seat bottom-out frequency', while the VDV model explained 19.5% of the variance in the directly measured VDV vector sum, with 11.6% described by the same five predictors as the RMS model. Predictive ability of the RMS model among 1000 bootstrapped samples can be anticipated to range from 14.3 to 69.1%, which may be considered adequate as exposure assessment tool for uses of epidemiological studies. The percentage of variance explained ranged from 0 to 40.5% for the VDV model, which is not robust and therefore likely not appropriate for use in survey-based exposure prediction. Whole-body vibration exposure modeling remains valuable, but is challenging in farming; the described model variance may increase with a more comprehensive list of candidate variables collected and quantified at machinery, farm, and farmer level. Predictors identified in the current and future models may provide a better understanding of how whole-body vibration exposure is modified, guide farmer's future decision on updating equipment, and allow for the development and initiation of interventions.