In this paper, we present some of the issues encountered when trying to apply model-driven approaches to the engineering of real-time systems. In real-time systems, quantitative values of time, as reflected through the duration of actions, are central to the system's correctness. We review basic time concepts and explain how time is handled in different modeling languages. We expose theinherent paradox of incorporating quantitative time-dependent behavior in high-level models. High-level models are typically built before the system is implemented, which makes quantitative time metrics difficult to predict since these metrics depend heavily on implementation details. We provide some possible answers to this paradox and explain how the Timed Abstract State Machine (TASM) language helps address some of these issues.