Abstract: Minimizing the energy consumption is an important design metric in embedded systems that are used in portable devices. However from a user's perspective, maximizing the operational lifetime of the device is even more important. This dissertation presents system level energy management policies that maximize the operational lifetime by considering the characteristics of both the embedded system and the power source. The first part of the dissertation deals with system-level energy minimization for embedded systems that support dynamic voltage scaling (DVS). While DVS effectively reduces dynamic power, in scaled CMOS technologies, static power plays an important role and DVS does not necessarily result in the reduction of the total power (sum of dynamic and static power). The proposed dynamic task scheduling and voltage scaling algorithms take into account both the dynamic power and the static power and minimize the system-level energy consumption. The second part of the dissertation deals with embedded systems that are powered by fuel cell-battery (FC-B) hybrid sources; the FC provides high energy density and the battery provides high power density. The optimization objective is no longer minimization of the energy consumption of embedded system but minimization of the fuel consumption of the power source. First, DVS-enabled system is considered and FC-aware algorithms are proposed for the case when the FC works at fixed output level and the case when the FC works at multiple output levels. The latter case achieves better performance since both the embedded system and the FC can be controlled. Next FC-aware algorithms are proposed for embedded systems that support dynamic power management (DPM). For such systems, the FC output setting that minimizes fuel consumption is formally derived, and this setting is applied on top of a prediction-based DPM policy to generate FC-aware algorithms. Finally an FC-aware energy management policy is proposed for a system which includes both DVS and DPM components. It is shown that this algorithm achieves higher fuel savings compared to algorithms which utilize only DVS or only DPM capabilities.