In this paper, we address a power-aware scheduling algorithm for mixed real-time tasks. A mixed-task system consists of periodic and sporadic tasks, each of which is characterized by its worst-case execution requirements and a deadline. We propose a dynamic voltage scaling algorithm called DVSMT, which dynamically scales down the supplying voltage (and thus the operating frequency) using on-line slack distribution when jobs complete earlier while still meeting their deadlines. Simulation results show that DVSMT saves up to 60% more than the existing algorithms both in the periodic and mixed task systems.