This is not a bad explanation but when doing this practically it can be useful to take log() of the probabilities so that you work with sums of logs rather than multiplying small floats.

http://getpopfile.org/docs/faq:bayesandlogs

From the article:

    For an English spam classifier that
    considers all the words in the English
    language, the number of the words (n)
    is approximately 171,476.

That's a remarkably precise number to be preceded by the word "approximately".

Nice article, very glad to read it. Keep up the good work.