When HP exited the optical character recognition business in 1995, its Tesseract OCR engine was released to UNLV as open source. In January, developers (including some from Google) decided Tesseract was stable enough to “re-release” as an open source project.
Don’t expect too much. You’ll need to compile the source code (if you use Cygwin, be prepared to copy /usr/include/limits.h to /usr/include/linux/limits.h or fix the source). And then you get a command line tool that reads single-column TIFF files from the command line. But the accuracy is much better than most of the cheap OCR tools out there.