We develop a novel transmit power-allocation strategy for wireless multimedia communications using cross-layer optimization. Because the individual frames in a compressed video carry different degrees of importance in terms of quality of service (QoS), improving the packet-loss probability does not necessarily coincide with perceivable improvements in QoS from a client's point of view. We propose a channel power-allocation scheme that properly incorporates the importance of frames when allocating transmit power. The system is modeled in terms of the number of remaining packets to be sent, the interference level, and the QoS importance of the packet that is being transmitted. Dynamic programming formulation is developed to obtain the optimal transmit power that simultaneously minimizes power consumption and maximizes the user-perceivable QoS. Using dynamic programming, we establish a state transition equation in discrete time intervals that incorporates the packet-loss probability. A simulation-based experiment is performed to test our transmit power-allocation scheme. We use four publicly available Motion Pictures Expert Group 4 (MPEG-4) video clips, comparing the performance of three different channel power-allocation strategies: constant power-allocation, interference-aware power-allocation, and significance-aware power-allocation. Optimal channel power under varying interference levels and the respective packet loss behaviors are examined. The results prove that our significance-aware transmit power-allocation scheme exhibits superior performance in terms of the loss ratio PSNR and power-utilization efficiency. The total packet byte loss decreases by 3%-8%. The loss rate of the I frame decreases from 24% to 1%. Significance-aware power-allocation successfully provides unequal protection to the more important packets. Total power consumption is reduced by 18%-34% without loss of user-perceivable QoS.