Email system since its inception has been a data exchanger between users over the internet. Email spam is the spoiler of the email system and a major concern towards the security and privacy related performance. Since Email Spam has to keep at lowest many researchers have given their definitions to handle this spam. The spam datasets used in the spam filtering area of study deal with large amounts of data containing irrelevant and/or redundant features. This redundant information has a negative impact on the accuracy and detection rate of many methods that have been used for detection and filtering. In this paper we propose a statistical feature selection approach combined with similarity coefficients are used to improve the accuracy and detection rate for the spam detection and filtering and prove the stability of detection rate, accuracy and false raising system.