In collaborative data mining, data sets from various parties are submitted to a third party where they are combined, privacy of each data sets’ sensitive attributes are protected and data mining is carried out. In privacy preserving data mining, there is a need to extract knowledge from databases without disclosing information about individuals. Each participant will have sensitive and non-sensitive data in their local database. Therefore the most important challenge in privacy preserving multi party collaborative data mining is how these multiple parties conduct data mining without disclosing each participant’s sensitive data. In this paper we propose a two-level encryption algorithm for protecting sensitive attributes from disclosure and a generalization algorithm. This approach guarantees high level privacy with less amount of complexity as compared to the existing methods and also proves to be fast and efficient over dynamic queries.