Ever wonder how good the analysts are at estimates? Ask the graders at StarMine. The quarterly earnings consensus, as delivered by Thomson Financial/First Call, is a powerful number. First Call lumps together analysts' forecasts for upcoming quarters and averages them. If a company "misses its number," as the portentous phrase goes, the market delivers medieval punishment. If it beats the number, Wall Street gives it a hero's welcome. But the First Call consensus has limitations. While analysts are justifiably knocked for sticking together like magpies, the range of estimates can be quite wide. The latter is especially true now that companies are forbidden to whisper hints in analysts' ears without disclosing the information to the public. Either way, taking an average of a groupthink cluster or of a wild array of calls creates an artificial consensus number that often can be plain wrong. Of the 4,695 companies reporting in the third quarter, the First Call consensus was off by 1 cent or more on 1,559 of them. "[The consensus is] the busy investor's equivalent of CliffsNotes, and that's being unfair to CliffsNotes," says Fredric E. Russell, an Oklahoma City money manager. Given how a stock can gyrate after an earnings surprise, having an inkling that one is upcoming can be profitable for investors willing to take a flier. Enter StarMine, a two-year-old Web site that offers its own rival consensus that you can use to flag upcoming surprises: The number indicates a surprise if it differs from First Call's. Of 237 StarMine earnings-surprise predictions published in FORBES since September 2000, the site has correctly picked whether a given company would exceed or miss the consensus 183 times. You can use the service for free by logging on to www.starmine.com. (First Call also is available through a free Web site.) How does StarMine arrive at its prognostications? By assessing the analysts who generate the First Call numbers. StarMine evaluates their forecasting skills by looking at the numbers collected since 1983 by First Call and by IBES (now swallowed by First Call). It ranks 3,650 analysts, awarding them one to five stars. Then it gives extra weighting to the four- and five-star analysts and arrives at its own estimate. To get five stars an analyst has to be closest to the mark and faster to get there than 90% of his peers. In ranking analysts, San Francisco-based StarMine echoes publications like Institutional Investor, as well as the Web outfit Bulldog Research (FORBES, Oct. 16, 2000). Bulldog, dependent on ads, growled its last in May, another victim of the dot-com wipeout. While StarMine isn't profitable, it at least has a steady source of income-selling an enhanced version of its service to big investors such as American Century. Annual fees range from $16,000 to $150,000, depending on the level of data required about analysts' records. Like Bulldog, StarMine seeks not to offend too much: You won't find any mention of clumsy analysts on the free site; only the paying customers can see that. And StarMine obligingly doesn't track analysts from Goldman Sachs, Merrill Lynch or Morgan Stanley on its free site. They deign to be carried only on the subscriber version. To StarMine, the prized analysts are those willing to make bold calls that signal things are worse than everyone's expectations. Like Douglas Lee, a five-star analyst for Banc of America Securities who began Micron Technology's fiscal fourth-quarter (June through August) calling for a 30-cent loss, when the First Call number was -19. Lee's call pulled the StarMine estimate down to -26. By August Lee had lowered his forecast to -40, dragging StarMine's to -29; First Call had dropped just a penny, to -20. The company ended up reporting -96. StarMine founder Joseph Gatto came up with the idea for the service as a frustrated investor. In 1995 he had a yen to buy shares of Dell Computer but was dismayed by the wide range of IBES numbers. Then the owner of a consulting firm that assessed corporate research and development plans, Gatto asked an IBES salesman how to find who the good analysts were. The guy said: "Well, you kind of just know." Because a big part of StarMine's cachet is showing up First Call, you'd think there would be ill will between them. Not so. Reason: The two have a business relationship. StarMine serves as a First Call distributor, meaning it sells that service along with itself and shares the First Call subscription revenue. Even though StarMine's consensus is implicitly critical of his service, First Call research director Charles Hill says the two offerings are "complementary." He can afford to look the other way. First Call's parent, Thomson Financial, has 400 times more revenue than StarMine, and First Call's paying service offers a vast amount of analysts' research to clients, which StarMine doesn't do. Besides, says Hill, the First Call estimate is still "what the market as a whole is keying off of." He's not wrong.