Objectives: Cancer survival information is available only in areas covered by cancer registration. The objective of this study is to project cancer survival for the entire US as well as states from survival data from the National Cancer Institute's Surveillance, Epidemiology, and End Results (SEER) program. Methods: Five-year breast, prostate, and colorectal cancer relative survival rates from SEER are regressed on socioeconomic, demographic, and health variables at the county level. These models are first validated by comparing the observed rates with projected rates for counties not used in the estimation process. Results: Education was the best indicator of longer cancer survival. Other important predictors of the geographical variability of survival varied by cancer site. Better survival was predicted for breast and prostate than for colorectal cancer. Conclusions: Data from cancer registries can be used in ecological models to provide national and state estimates of patients' survival rates. These estimates are useful in targeting areas in which to promote earlier diagnosis or improved access to care, and may also aid in monitoring the quality of survival data collected by individual cancer registries.