Validation of dynamics model prediction is challenging due to the involvement of various sources of uncertainty and variations among validation experiments and over time. This paper investigates quantitative approaches for the validation of dynamics models using fully characterized experiments, in which both inputs and outputs of the models and experiments are measured and reported. Existing validation methods for dynamics models use feature-based metrics to give an overall measure of agreement over the entire time history, but do not capture the model's performance at specific time instants or duration; this is important for systems that operate in different regimes in different stages of the time history. Therefore, three new validation metrics are proposed by extending the model reliability metric (a distance-based probabilistic metric) to dynamics problems. The proposed three time-domain model reliability metrics consider instantaneous reliability, first-passage reliability, and accumulated reliability. These three reliability metrics that perform time-domain comparison overcome the limitations of current feature-based validation metrics and provide quantitative assessment regarding the agreement between the simulation model and experiment over time from three different perspectives. The selection of validation metrics from a decision-making point of view is also discussed. Two engineering examples, including a simply supported beam under stochastic loading and the Sandia National Laboratories structural dynamics challenge problem, are used to illustrate the proposed time-domain validation metrics.