BACKGROUND AND PURPOSE: Of the 795,000 strokes occurring in the USA each year, over 20% are recurrent events. Little is known about how the rates of recurrent stroke in the country have changed over time. Our objective was to determine national trends in 1-year recurrent ischemic stroke rates by US county among the elderly from 1994 to 2002. METHODS: One-year recurrent stroke rates following incident ischemic stroke (ICD-9 433, 434, 436) among all fee-for-service Medicare beneficiaries were determined by US county for 1994-1996, 1997-1999, and 2000-2002. Bayesian spatiotemporal Poisson modeling was used to determine county-specific trends in recurrent stroke rates over time with risk adjustment for demographics, medical history and comorbid conditions. RESULTS: The analysis included more than 2.5 million beneficiaries (56% women; mean age: 78 years; 87% white; n = 957,933 for 1994-1996; n = 838,330 for 1996-1999; n = 895,916 for 2000-2002) aggregated to all 3,118 US counties. After adjustment for changing patient demographics and comorbidities, there was a 4.5% decrease in recurrent stroke rates from 1994-1996 (13.2%) to 2000-2002 (12.6%; p for trend <0.0001). The geographic and temporal patterns were not uniform; the recurrent stroke rates decreased within sections of the Southeast (the 'stroke belt'), but increased in counties in the middle and western sections of the USA. CONCLUSIONS: The overall recurrent ischemic stroke rates declined by almost 5% from 1994 to 2002, but temporal patterns varied markedly by region. Additional research is needed to identify the reasons for this geographic disparity.