Given an arbitrary set of symbols (the english alphabet is the example that will be used here), Huffman coding is a way of creating the most efficient (smallest) binary code for that set of symbols. It exploits the fact that in most data some symbols occur more frequently than others and by shortening the code for that symbol, space can be saved. There are other interesting properties of Huffman codes as well as some gotchas with respect to their use, so the above link is worth reading.