Large-scale health surveys are major data sources for health policy research, and telephone survey is the major mode for national or state health surveys. With the dramatic increase of cell phone use over the past decade, large telephone surveys need to employ the dual-frame design, which greatly improves the representativeness of the sample. From a practical perspective, it is not clear how to optimally implement the dual-frame sampling design. This article studies the impact of different design and analytical strategies of dual-frame surveys on the estimation of population quantities via both simulation studies and using data from the Ohio Family Health Survey (OFHS). Results imply that cell-only design is less biased than the cell-any design with no or poor post-stratification. If accurate post-stratification information is available for different phone use patterns, all estimation techniques provide similar results and the cell-any supplemental sampling is more cost efficient. Practical recommendations for health policy researchers are provided based on applying different analytical methods to the OFHS data.