In 2013, the U.S. Census Bureau added an option to respond to the American Community Survey online. With this addition, the mailing strategy was changed to push internet response in the first mailing. The option to respond by a paper questionnaire was available in the third mailing. As a result of this strategy change, self-response rates decreased in areas believed to have low internet availability or a preference to respond by mail. In 2017, a test was fielded where a targeted sample of addresses were offered a choice to respond by either internet or paper in the first mailing. To decide which households received a choice in the first mailing, we used a classification algorithm to categorize census tracts into one of three groups: Mail Preference (19% of tracts), Mixed Preference (12% of tracts), and Internet Preference (69% of tracts). Roughly half of the addresses in the Mail Preference and Mixed Preference tracts were sent materials that offered a choice in the first mailing while the other half of the addresses were offered a choice in the third mailing (the current mailing strategy). This presentation will describe the classification algorithm and will include some preliminary results comparing these two mailing strategies.