It’s entirely possible that for all the talk about the gender gap, Donald Trump could prove more competitive among white women against Hillary Clinton than it appears today.

But Trump faces at least as much risk that white women could seal his doom in the general election match-up against Clinton that now appears certain.

How could both things be true? The answer is that the electorate’s changing composition makes it virtually impossible for Trump to prevail in November unless he not only wins most white women, but also carries that group by a convincing margin. Even a narrow lead for Trump over Clinton among white women would likely ensure his defeat.

The good news for Trump is that the Republican edge among white women has widened substantially since 2000, providing him a foundation on which to build. The bad news is he faces enormous, possibly unprecedented headwinds in defending that advantage. “There is something really basic, elemental, going on here with women reacting to [Trump],” said the Democratic pollster Stanley Greenberg, who has polled extensively for the Women’s Voices Women Vote Action Fund, a liberal voter mobilization group. “Every signal he’s sent that has built up his support with men and Republicans has had the opposite effect with women.”

The widespread discussion about the gender gap can obscure more than it illuminates. Since the 1980s, Democratic presidential candidates have consistently run better among women than among men—or put the other way, Republicans have run better with men than women. This pattern has held among whites, African Americans, and Hispanics.

But the magnitude of the gender gap is inflated by Democrats’ overwhelming advantage among women of color. President Obama, for instance won exactly 96 percent of African American women, and over two-thirds of Hispanic women, in each of his two victories. Democrats haven’t done nearly as well among white women. In modern exit polling tracing back to 1972, the only Democrat to win more white women than his Republican opponent was Bill Clinton in 1996. Clinton in 1992 and Al Gore in 2000 also ran about even with white women. But since then, the GOP has carried white women by solid margins: 11 points in 2004, seven in 2008, and fully 14 for Mitt Romney against Obama in 2012.

Romney last time carried white men by an even more emphatic 27-point margin (the Republicans’ best showing since 1988). Yet because of his crushing advantage among the growing minority population, Obama still won reelection by 5 million votes.

That precedent frames the challenge facing Trump, now the inevitable GOP nominee. Following long-term trends, it’s likely that minorities will cast 30 percent of the 2016 vote, and white women will comprise slightly more of the remainder than white men—just as in 2012. In his absolute best-case scenario, Trump might match the two-thirds of white men that Reagan won in 1984, the party’s modern apex. But given Trump’s astronomical unfavorable ratings among African Americans and Hispanics, it’s not unreasonable to project that Clinton could hold the roughly 80 percent of minority voters who have typically backed Democratic nominees since 1976.

If both those projections held true, and the electorate’s composition followed the long-term patterns, Trump would then need to attract 58 percent of white women to reach a national majority—slightly more than the 56 percent that Romney won. Looking at the equation from the other direction, if Clinton matched the usual Democratic performance with non-white voters and also carried even half of white women, Trump would then need to win more than three-fourths of white men for a national majority, a daunting prospect.

These calculations lead to the same place: Trump almost certainly can’t beat Clinton, or even stay competitive, without constructing a solid advantage among white women. But today he’s trailing Clinton among them in most surveys. Single and college-educated white women usually lean Democratic, but Trump in early polls faces unusually large deficits with them. (The Democrats’ biggest advantage among college-educated white women since 1980 has been eight points; a compilation of all NBC/Wall Street Journal national polls this year showed Clinton leading Trump among them by 21 points.) That resistance may be extremely difficult to crack, since, as Greenberg notes, it’s rooted less in policy than a visceral sense that Trump represents an “assertive traditional [male] attitude toward women” that rejects the gender revolution that has recast relations between men and women in the home and workplace alike.

Married and non-college-educated white women could provide Trump more opportunity. Both groups have preferred the GOP by double-digit margins in each election since 2004. These women tend to be more culturally conservative, and they may also be more receptive to Trump’s confrontational messages on the economy and terrorism. And in some polls, non-college–educated white women in particular also view Clinton about as unfavorably as they do Trump. “They are struggling every day; they have a very difficult life,” Karlyn Bowman, a public-opinion analyst at the free-market-oriented American Enterprise Institute said. “They don’t get to go to speeches at Goldman Sachs.” Yet polls today generally show Trump lagging behind Romney’s 20-point edge with these women. (The NBC/WSJ compilation actually shows Clinton leading among them, though other recent surveys merely give Trump a narrower advantage than Romney.)

Gains among blue-collar white women won’t save Trump if he loses their white-collar counterparts by the margins polls now suggest. But Trump needs to start somewhere if he’s to remain competitive in the fall. In the weeks ahead, his standing among the economically-strained working-class white women also cool to Clinton may be the most revealing indicator of whether Trump has any chance to emerge from the general-election ditch he’s dug in his bombastic march toward the GOP nomination.