Long-term exposure to elevated radon (222Rn) concentrations has been linked to increased lung cancer risk. Year-long measurements of contemporary radon concentrations have been the "gold standard" for epidemiologists trying to reconstruct past radon exposures and for homeowners trying to estimate future radon exposure. Random variations and persistent temporal trends can affect remedial action decisions and risk coefficients derived from epidemiological studies. Temporal fluctuations are possible when changes occur in a home's structure, climate, environment, or occupants. The annual-average temporal radon behavior was studied at 196 sites in 98 Minnesota houses. Seventeen hundred year-long indoor radon measurements were made from 1983 to 2000 to determine year-to-year radon fluctuations and long-term temporal trends. Ten year-long measurements over a span of 13 years were made at the typical site. The median radon concentration was 120 Bq m-3. The median radon concentration of the group of houses showed little year-to-year variation and no persistent temporal trends. At individual sites, year-to-year radon variations ranged from 3 to 110%. The median variation was 26%. Climate, exposure to wind, and radon concentration affected year-to-year variation, but house age, construction, or measurement floor did not. Some individual sites showed significantly larger radon changes when modifications were made to the house structure and heating-ventilation systems. Year-long radon measurements on the first floor provided better estimates of cumulative radon exposure than screening measurements. The radon variations observed in this study provide uncertainty estimates for year-long measurements that could help improve remediation decision protocols and refine risk estimates from epidemiological studies.