Each word within an e-mail is assigned two word counts:
From these values, a spam probability is computed for the word. This setting examines the words whose probabilities deviate most from an average word. These words are both spam and non-spam words.
Example: Suppose this option is set to 15. Since most words have an average spam probability of 50% (50% likely to be spam, 50% likely to be non-spam), then the fifteen words that are furthest away from 50% are used. So if a word has a spam probability of 5% it will most likely be used. Likewise, if a word has a spam probability of 90%, it will most likely be used. A word that has a 45% probability will most likely not be used.