The Census Bureau’s American Community Survey (ACS) is planned to be an annual sample of three million housing units mailed out in monthly panels. Data collection for each monthly panel extends over a three-month period: mailout/mailback in the first month, telephone follow-up in the second month for addresses where a telephone number can be obtained, and personal-visit followup in the third month for a onethird subsample of the remaining nonrespondents. Areas with low mail response will have a larger percentage of cases going to personal-visit, and thus will have a larger variance estimate because of the subsampling in this phase. This paper will examine several options of oversampling these areas to increase the reliability of the estimates.