As response rates for RDD surveys and the household coverage of landline-only RDD sample frames continue to decline, population estimates from RDD surveys are increasingly vulnerable to both nonresponse and noncoverage bias. Yet, RDD remains the most cost-effective approach to surveying individuals in the general household population; mail and Web frames and survey methods have not yet proven to be viable substitutes. Several RDD surveys have experimented with supplementing the landline sample with either area frames or frames of telephone numbers reserved for cellular service, in order to assess nonresponse and noncoverage bias. The 2007 California Health Interview Survey is implementing both an area sample in Los Angeles County to assess nonresponse bias and statewide samples of landline and cellular telephone numbers to produce estimates for individuals in households with telephones. While these kinds of enhancements may improve RDD-based estimates, the per-case costs of interviewing from area and cell phone frames are typically higher than from landline RDD frames. This paper will draw upon the CHIS 2007 experiences to compare the relative costs of interviewing from landline, cellular, and area sample frames in a mixed-mode design. It will consider interviewer time and related expenses; supervisory, management, and processing costs; the marginal costs of adding a frame or mode. Finally, it will explore the optimal allocation of sample to frames with various assumptions about the contribution to variance of each frame.