This is a bug that affected Intel Pentium chips between 60 - 100 MHz before 1995.
The problem meant that there was a 1 in 360 billion chance of there being an error in the 4th significant figure of a decimal number and 1 in only 9 billion chance of an error in the 10th significant figure. To check if your processor is affected, perform the operation 4195835/3145727 in the Microsoft desktop calculator for an example of the problem. The result should round to 1.3338, not 1.3337.
The error was caused by a some omissions (5 to be precise) in a look up table of 1066 entries used to speed up the process of floating point division (FDIV).
The biggest problem wasn't with the bug, it was with Intels refusal to acknowledge it as a bug. At first, they denied that it existed. When this was finally proved (they now refered to it as a flaw and not a bug), users then had to prove that they were affected by this bug before they were entitled to a replacement. In the end, after massive public outcry, Intel agreed to replace all affected chips unconditionally.
Sources
http://www.ukans.edu/cwis/units/IPPBR/pentium_fdiv/pentgrph.html
http://www.cs.earlham.edu/~dusko/cs63/fdiv.html
http://x86.ddj.com/errata/feb97/bugs.htm
http://support.intel.com/support/processors/pentium/fdiv/