Markov models are used in many industrial applications, but, for very large models, simulation is often currently the only viable evaluation technique. However, simulation techniques that are based on evaluating trajectories at the level of individual states and transitions can be inefficient because they have to keep track of many details. Moreover, since they use statistical methods, estimating solutions at higher confidence intervals requires the evaluation of an increasingly large number of trajectories which often leads to poor performance. On the other hand, analytical path-based techniques can be used for computing guaranteed bounds on the true solutions, but they can have poor performance because they must evaluate many paths to obtain reasonable bounds. In this paper, we present a path-based simulation approach for evaluating models at the component, rather than individual state/transition, level. At this level of abstraction, the approach can compute more accurate solutions than traditional discrete-event simulation techniques can in a given amount of time. In addition to presenting the approach, we compare its performance and effectiveness against a path-based analytic technique.