The main aim of project is to develop a system which identifies and detects fraud in online credit card transactions accurately. The system follows a multi-layered approach for security based on the amount of the transaction. We have used the probabilistic algorithm, an improvement on the Hidden Markov Model to detect frauds more efficiently. The algorithm calculates a threshold value based on the previous transactions and classifies them into 3 categories-low, medium and high. The current transaction is compared with the threshold value and on the basis of the calculated probability it is classified as fraud or not. The system also implements certain authentication mechanisms to detect the fraud in real time and block the transaction if the user is not found to be legitimate.