Spam Hound uses Bayesian Filtering to detect spam.

Bayesian filtering performs a statistical analysis of the words that make up the email to determine if the email is spam or a personal email.

Spam hound combines Bayesian filtering with the other spam detection techniques using a state of the art Neural Network which learns to accurately detect spam based on your personal email usage.

Spammers try to defeat this technique by adding random text (e.g. "Tables bark deliciously...") to their spam.
This is also why some spam email contain only an image with the text printed on it.
Spam filters which rely solely on Bayesian filtering will miss these spam.
Bayesian filtering does however remain effective in detecting personal email since spammers are unlikely to know your personal interests and so email that contain words relating to your interests are easily detected and allowed through.