Tests of income convergence yield mixed results and omission of spatial effects might be one cause. This paper examines the impact of geographic aggregation and spatial effects on conditional convergence in the United States from 1970 to 2004 at three levels of aggregation. A standard ordinary least squares (OLS) conditional convergence model is first developed. Model diagnostics, however, suggest that a spatial autoregressive (SAR) model is appropriate. OLS and SAR models are compared across scales according to their model fit, diagnostics, convergence evidence and possible correlates. Results indicate that (1) convergence evidence and models are sensitive to spatial effects; (2) spatial models consistently outperform OLS; (3) model fit is better at larger aggregations, while convergence evidence is strongest at smaller aggregations; and (4) spatial aggregation needs to be an explicit concern in the construction of future convergence models. Factors driving convergence remain generally consistent through impact varies by scale.