Administrative income tax data indicate that U.S. top income and wealth shares are both substantial and larger than shares observed in household surveys. However, these estimates are sensitive to the unit of analysis, the income concept measured in tax records, and, in the case of wealth, to assumptions about the correlation between income and wealth. We constrain a household survey--the Survey of Consumer Finances--to be conceptually comparable to tax records and are able to reconcile the much of the difference between the survey and administrative estimates. Wealth estimates from administrative income tax data are sensitive to model parameters.