We examine the relative performance of voting technologies by studying presidential, gubernatorial, and senatorial election returns across hundreds of counties in the United States from 1988 to 2000. Relying on a fixed-effects regression applied to an unbalanced panel of counties, we find that in presidential elections, traditional paper ballots produce the lowest rates of uncounted votes (i.e., “residual votes”), followed by optically scanned ballots, mechanical lever machines, direct register electronic machines (DREs), and punch cards. In gubernatorial and senatorial races, paper, optical scan ballots, and DREs are significantly better in minimizing the residual vote rate than mechanical lever machines and punch cards. If all jurisdictions in the United States that used punch cards in 2000 had used optically scanned ballots instead, we estimate that approximately 500,000 more votes would have been attributed to presidential candidates nationwide.