Four years ago, Al Gore got half a million more votes than George W. Bush -- about one-half of one percent of the total -- but, thanks to Florida, Bush won the electoral vote. Democrats have been outraged ever since. What would happen if Bush or Kerry were to win the popular vote by three or four million votes -- but still lose in the electoral college? Welcome to the Mother of All Legitimacy Crises. And to the Administration That Cannot Govern.

It could easily happen. Based on the polls reported at www.realpolitics.com, Bush is running well ahead of his 2000 performance in the I-95 corridor. (Michael Barone wrote a wonderful column about this phenomenon a couple of weeks ago.) Kerry will still win Maryland, New Jersey, New York, and Connecticut. But he will win them by smaller margins than Gore did. If Bush improves on his 2000 performance in the South and West -- easily possible, given that a New Englander and not a Tennesseean is heading the Democratic ticket -- he could pile up a margin of a few million votes, and still lose Ohio and Iowa and with them the election.

Democrats would call that poetic justice, and maybe they're right. But it can't be good for the country to have something America has never seen before: two consecutive Presidents who lost the popular vote. One of them by a lot.

We have come closer to this particular train wreck than people think. In 1896, William Jennings Bryan lost the popular vote to William McKinley (Karl Rove's second-favorite President) by more than four percentage points. But a shift of 20,000 voters -- about one-seventh of one percent of the total -- in six states would have given Bryan an electoral-college victory. Woodrow Wilson won the popular vote in 1916 by more than three percentage points, the equivalent of a three-million-plus vote margin today. But if Charles Evans Hughes had persuaded 1,900 more Wilson voters in California -- two-tenths of one percent of that state's vote -- he would have won the White House. In 1948, Harry Truman ran four-and-a-half percentage points ahead of New York Governor Thomas E. Dewey; a comparable margin today would be nearly five million votes. Change 3,600 votes in Ohio and 9,000 in California, and the 1948 election goes to the House of Representatives. Change another 17,000 votes in Illinois, and Dewey wins outright. Each of these changes represented less than half of one percent of the relevant states' votes.

In 1976, it almost happened again. Jimmy Carter beat Gerald Ford by 1.7 million votes nationally, two percent of the total. Change 5,600 votes in Ohio and 7,300 in Mississippi, and Ford is the one walking down Pennsylvania Avenue on January 20, 1977.

Of course, all these near-misses are still misses. Not since 1876 has a candidate won the popular vote by as much as a full percentage point while losing in the electoral college. But that pattern is likely to be broken -- if not this year, then soon. And often.

Consider an important feature of all the elections mentioned above. Neither Bryan nor McKinley knew which six states would decide the White House in 1896. (In case you're wondering, they were West Virginia, Kentucky, Indiana, North Dakota, Oregon, and California.) Neither Wilson nor Hughes knew that California would decide the contest in 1916, anymore than Truman or Dewey expected their election to turn on Ohio, Illinois, and California. Even in 1976, the art of opinion polling was sufficiently imprecise that neither campaign was likely to guess that Ohio and Mississippi would hold the keys to the kingdom.

That ignorance was good for American democracy. It meant that candidates had to run national campaigns. To be sure, for most of our history Republicans could ignore the South, which regularly rolled up large Democratic majorities. And most years, Democrats could ignore Republican strongholds in New England and the Upper Midwest. (Times have changed; the two parties' geographical bases have switched sides.) But no presidential candidate could afford to focus all his energies on a handful of "battleground states" -- those states existed, but no one could know in advance where they were.

Today, Karl Rove knows. So do Bob Shrum and Mary Beth Cahill. Polling is miles better and more sophisticated than it was even a generation ago, when Gerald Ford almost snuck by Jimmy Carter. Which means that America is not really selecting the President on November 2. Ohioans and Iowans and New Mexicans are. (I may not have the right three states, but you can bet that the campaigns do.) And if one of these two campaigns makes better, more targeted investments in the right two or three states, that campaign will carry the day -- even if millions more Americans vote for the other side.

Two other changes in the political landscape make that scenario likely. As recently as 1960, forty-five percent of the voters cast their ballots in states decided by three points or less. In 2000, the number was fourteen percent. In 1960, Kennedy and Nixon won nine states by fifteen points or more. In 2000, Bush and Gore won twenty-two states (plus the District of Columbia) by margins that large. Safe states, once the exception, are now the rule. Swing states, once common, are few. The second change involves advertising. The rise of cable makes it easy for candidates to speak to small slices of the electorate. Campaigns can focus their attention, time, and money where those things will do the most good -- and write off large chunks of the country. The odds of a Bryan or Dewey winning the election while losing the popular vote by several percentage points are much higher as a consequence.

That isn't good for American democracy. We've always known that the Electoral College allows minority presidents. But today, it virtually guarantees them. Over time, as polling and communication become more precise, candidates' electoral vote totals will correlate less and less with the nationwide popular vote. We could have a string of seemingly illegitimate presidencies.

Plainly, the machinery is broken and needs fixing. The Electoral College could still serve a useful role -- with two changes. First, get rid of the electors: human electors are a disaster waiting to happen, an invitation to bribery, fraud, or simple stupidity. Just give each candidate the number of electoral votes he earns on election day. Second, make each state do what Nebraska and Maine do: give one vote to the winning candidate in each congressional district, and two votes to the candidate who wins the state. It's not perfect, but it's a pretty good way to make sure that Presidents have broad, geographically diverse support without having the election turn on one or two states. It would work even better if the Supreme Court -- where are activist judges when you need them? -- would ban partisan gerrymandering, so we could have more close districts.

America was lucky to escape Presidents Bryan, Hughes, and Dewey. But our luck is running out. We need to change the system, and soon.

In the meantime, I'm rooting for a landslide.

William J. Stuntz is a Professor at Harvard Law School.