In late years, with the spread of Internet, increase of an unwanted e-mail becomes a problem. In 2006, there is a finding that spam-mail is occupied 91% of the whole E-mail. A Bayesian filter filtering statistically with Bayes theory as an anti-unwanted e-mail measure is used widely. The filter has high precision of filtering and is able to match with the trend of an unwanted e-mail and personal preference as the characteristic. However, the image spam that does imaging of contents of an unwanted e-mail increases rapidly as the end run. Baysian filter is not able to do an appropriate learning and judgement for binary data such as images since Bayesian filter learns and judges for only text data. Therefore, in this paper, we pay attention to information, such as file size of an attached image and suggest a technique of filtering with adding these information to a corpus of existing Bayesian filter as an anti-image spam measure. In addition, we report the evaluation result.