In this paper, we propose a design methodology for the inter-prediction in H.264/AVC codecs by addressing the relationship between its main processes. The target of this methodology is to optimize the design in order to get better performance while keeping a reasonable design cost. An efficient hardware architecture for the inter-prediction in H.264/AVC codecs is then proposed with three key techniques: a modified full search algorithm with bandwidth efficiency, pipelining technique, and data reuse strategy. With this approach, the inter-prediction has been successfully designed and implemented with a CMOS 180nm technology which provides low cost in terms of latency, hardware overhead and memory bandwidth. The design is initially targeted to CIF video format; however, it is obviously suitable for real-time HD 1080p video format.