You can read an informal explanation onhttp://www.tedpack.org/gedcom.html
Basically, your program/web site exports (writes) a bunch of 80-character records to a text file named something.GED, and then your other program/web site imports (reads) them at 10,000 lines a second, which is faster than you can type. 80 characters is left over from IBM cards, which are, in turn, left over from the days computers were powered by steam.
If you import into a program on your computer, the cards that the program can't read will go into a file named "filename.LST", in the same directory the GED file was, where "filename" is the name of the .GED file. You can then look at the mistakes and try to fix them. Most programs will write and read the standard facts; BMD dates and places. They get confused with esoteric stuff. Cause of Death used to be a problem, for instance.
If you can figure out who has the problem from the context ("COD Shot by jealous husband", for instance), fine; if not, you'll have to open both the GEDCOM and the LST file with Notepad, side by side, and see if you can figure out who has the unrecognized fact.