As states have expanded early and by-mail voting options, while often continuing to have election- day voting, the challenge of how to design a sample that accurately predicts the overall vote has become more complex. For more than three decades we have conducted a statewide exit poll designed largely to accurately predict the vote and provide data for academic analysis. In 2014 some more populous counties in our state opted to move to vote-by-mail and as a result we developed a sampling strategy that merged an election day sample design with a sample designed to learn about the voting behavior of early voters. One of the counties that shifted to a largely vote-by-mail system allowed voters to vote in-person on election day at voting centers. This county is part of two congressional districts which complicated data collection. In this county we experimented with I-pads as the means of data collection which allowed us to use zip codes to assign voters to the correct congressional district. This paper describes the methodology used in designing the sample and contacting the voters in a multimode survey approach. We contacted more than 20,000 people who had voted early by postcard asking them to go on-line and complete our survey. In addition we had 1,185 voters complete the survey over the phone. Non -response varied by mode of survey administration as did the sample size. Even with this added complexity the survey accurately predicted the closest congressional race in the state well within the margin of error. This paper describes what we learned in the process that may be of use to others who do exit polling of surveys using multimode designs.