On Thursday, the Fed, through its annual probe of bank stability, determined that among the nation’s top six banks, Citi (C), which essentially failed a year ago, was best positioned to weather another financial crisis. Goldman (GS) was second to the bottom on the main measure of capital, followed only by Morgan Stanley (MS). But on an alternative measure of capital that looks at all of the bank’s assets, and not just the ones generally deemed risky, Goldman came out looking the most vulnerable in a downturn. And not just compared to its close rivals but to all 18 banks in the Fed’s stress test.

A year ago, Citi was forced to resubmit its capital plan, erasing any proposal to raise dividends or buy back shares — two moves that investors like, but that deplete capital that could be needed to cover bad loans. This year, it was Goldman’s turn to be befuddled. Goldman released its own test of its finances using the same economic criteria as the Fed. Unsurprisingly, Goldman got better grades in its test. The firm is likely to spend the next week trying to convince the Fed why the regulator is wrong.

Most banks, including Goldman, are planning to request approval to increase their dividends or share buybacks, and next Thursday the Fed will officially approve or reject them. The results of the Fed’s test probably means that Goldman will have to curtail some of its planned increase in payouts. Shortly after the Fed released its results, Citi said it had asked permission to buy back $1.2 billion in stock in 2013.

MORE: Stress test results: Banks stand to lose nearly a half trillion dollars

This is all a very different view than the market has had of these two firms. Goldman’s stock is roughly where it was before the financial crisis. Citi’s shares are down 77%. What’s more, the market values Citi at 70% of its book value. Anything less than 100% is considered troubled. Goldman is valued at 110% of its book value.

One response to the Fed’s surprising results is that there is something wrong with its test. The other is that the stress test is picking up something that the market is missing, which after all is the point of doing stress tests.

For Goldman, the Fed seems to suggest that it is more exposed to a market shock than the firm has let on. In the notes to its latest financial fillings, Goldman said that it had cut the risk of its trading and investment book by 25% in the past year. The firm has said that it shuttered its proprietary trading unit, which makes bets with the firm’s own money.

But the Fed’s stress tests suggest Goldman has not done as good a job reducing risk in its trading operation as the bank suggests. In its stress test, the Fed said that Goldman’s trading operations could lose as much as $25 billion if there were another financial shock. That would erase more than three years of earnings and more than half of the bank’s capital.

MORE: How Goldman Sachs beat the Volcker Rule

Citi, on the other hand, seems to have done a much better job building up its defenses against a financial storm. In the past year, Citi has cut its operating expenses by $400 million, and pledge to cut a lot more. And it has upped the provisions it has put away for bad loans, to protect against future loans losses. The bank’s capital has risen $20 billion in the past two years.

At a minimum, the bank’s good stress test results may suggest that Citi ex-CEO Vikram Pandit, who was ousted in part because of the firm’s weak performance on last year’s stress test, deserves more credit for turning around the bank than he got. The Fed’s stress test results reflect financial results through the third quarter of last year. Pandit was forced out in mid-October. But his replacement Michael Corbat has announced new layoffs, and signaled that he will be more aggressive in cost cutting than his predecessor, which will further improve Citi’s capital ratios.

You could argue that the Fed went easy on Citi and the other big bank’s lending operations. In general, the Fed assumed lower loan loss rates for the big banks than it did a year ago, under the same economic scenario. The Fed, for instance, predicted that 9.2% of the Citi’s loans would go unpaid in its adverse economic scenario. That’s down from a prediction of 11.3% a year ago. Had the Fed stuck with its prediction of year ago, Citi would have $10 billion in additional losses. It’s not clear why the Fed reduced loan loss rates. A Fed official said that there was an assumption that banks are being more careful when they make loans these days. But the banks still have plenty of loans that were put on their books before the financial crisis.

MORE: Why the Fed is failing to boost lending

Still, the Fed’s stress test seem to be in line with the recent assessments of credit ratings agencies Moody’s and S&P, which have said that stand alone investment banks like Goldman and Morgan Stanley are likely to be riskier in the future than banks that rely more on lending. And while the financial crisis was mostly caused by bad loans, that’s not necessarily how the next one will play out.  Some are worried that the next financial shock will come from rising interest rates and big losses in the banks’ debt portfolios.

In the past, the Fed’s stress tests have been criticized for being too backward looking. This year, they appear to have done a little better job anticipating losses that the banks and markets don’t expect to happen. You could argue that makes the Fed’s stress test wrong or unrealistic. But that, again, is the point.