(FORTUNE Magazine) – Judging from the behavior of newly public Internet stocks, you'd think investment bankers haven't a clue about how to price an initial public offering. (Well, that may be partly true: Only psychologists and perhaps tarot card readers can guess exactly how rapid-clicking day traders will react to each new Web issue.)

Cases in point: Autoweb.com, the Santa Clara, Calif., online car-buying service, expected to sell 21% of the company at $10 to $12, ultimately offered those shares for $14 on March 23, peaked at $50 a day later, and thereafter dawdled in the low $30s (see "Who'll Be the Amazon.com of the $1 Trillion Car Biz?" April 26, in the fortune.com archive). San Francisco's Critical Path, an e-mail hosting service, first suggested offering a 13% stake between $9 and $11, then upped the projection to $22 to $24. The initial shares eventually sold for $24 on March 29 and shot over $90 before they were four days old.

Logic suggests, then, that companies going public are, in Wall Street's parlance, "leaving money on the table" by selling their initial stakes at too low a price. Investment bankers, for their part, could earn more fee income if the offerings were priced higher. Meanwhile, the giant institutions that buy the shares are getting too sweet a deal at the expense of individual investors, who rarely get a shot until the stock is trading higher on the open market.

The truth is somewhat more complex. Wily underwriters have a good sense of what's going on--and are profiting handsomely. Ditto for the companies selling stock. But pity the unsuspecting investor buying shares at market prices as Internet stocks shoot out of the gate, only to watch them falter soon after. "What's developing is a huge spread between the level where institutions are willing to buy for fundamental reasons and what buyers in the secondary market are willing to pay," says Scott A. Ryles, head of Merrill Lynch's technology banking group in Palo Alto.

By pricing shares lower than what the market might bear, investment bankers are protecting issuers from a fate worse than unrealized equity sales: the prospect that a sour market or problems with the company could drive shares below the IPO price. Companies that cause powerful institutions to lose money on an IPO won't find too many takers the next time they come to market.

Anyway, the companies typically tender a relatively small slice of available shares in the IPO, leaving room for improvement in the pricing of subsequent offerings. "You expect to raise only a small subset of the money that you want to raise over time," says Kenneth A. Goldman, chief financial officer of cable-based Internet service @Home. "So even if you 'leave some money on the table,' you will have happy shareholders and be able to raise more money down the road."

Indeed, @Home is a textbook example, though from a long-ago era of Net hype--two years ago. It raised about $100 million (after underwriters' fees) at $10.50 per share in its July 1997 IPO. That proved to be less than half what it theoretically could have gotten, since the shares opened near $25. But when the company returned to the capital markets 13 months later, it raised fully $126 million by selling far fewer shares--just 2.875 million, vs. 11.7 million the first time--at $46.13. (@Home stock recently fetched more than $185.)

There is an alternative for companies that feel the IPO process doesn't work--the so-called Dutch auction. This is a method whereby would-be investors submit bids, and the underwriter tabulates them to come up with an offering price. W.R. Hambrecht & Co., a new San Francisco investment bank started by Hambrecht & Quist co-founder Bill Hambrecht, is the first--and so far only--big proponent.

Establishment bankers scoff at the notion, undoubtedly in no small part because Hambrecht plans to take a commission of just 3% to 5% of the amount raised, rather than the typical 7%. But skeptics have more defensible arguments too. Sure, auctions may be fine today, when anyone, it seems, can hawk an IPO. In saner times, though, the system needs the underwriter's analysis and its ability to use its own capital to buy shares if necessary. Institutions, grouses Merrill Lynch's Ryles, "want judgment in the market. They don't want an algorithm to determine the price."

IPO pricing is working fine in the rest of the stock market. And eventually even Net stocks will look like any other business when it comes to financing.

ADAM LASHINSKY is a columnist for the San Jose Mercury News. You can read his column at http://www.mercurycenter.com/ svtech/columns/siliconstreet/index.shtml and e-mail him at alashinsky@sjmercury.com.