A (usually) binary encoding of the elements of a finite set, A,
where each element 
ai in 
A has an assumed probability, 
pi, of occurring in a message. The binary encoding satisfies the prefix property and is such that messages will have a minimum expected length. Thus an element 
ai with a high probability of occurring in a message is encoded as a short binary string while an element with a low probability of occurring is encoded with a longer string. 
See also source coding.