Two air conditioners in Minnesota were moved near a temperature sensor. But something didn't look right about the data. (When the NOAA's NCDC learned of examination of its network, it briefly removed information from the Internet.) McIntyre, a statistician, found that NASA does not make available the computer code and corrections used to "correct" its data. He had to reverse-engineer their calculations and found a year 2000 bug. GISS fixed their data and thanked McIntyre. It has been separately noted that 1934 was also warm in northern Europe too.
How significant are such "corrections"? To the 0.1C measured temperature increase, the NOAA adds 0.5C of adjustments.
Update: Related story posted: California Joins Open Document Bandwagon
It is easier to change the specification to fit the program than vice versa.