This paper implements a probability based method for fraud detection in telecommunication sector. We used Naïve-Bayesian classification to calculate the probability and an adapted version of KL-divergence to identify the fraudulent customers on the basis of subscription. Each user’s data corresponds to one record in the database. Since, the data involves continuous numerical values, the Naïve-Bayesian classification for continuous values is used. This methodology overcomes the problem of existing system, which classifies the best customer as fraudulent customers, as it works on a threshold based method.