ABSTRACTCorrelation power analysis (CPA) has been a powerful and thoroughly studied threat for implementations of block ciphers and public key algorithms but not yet for stream ciphers. This paper proposes a novel CPA attack on the hardware‐oriented stream cipher Trivium, one of the finally chosen ciphers by the eSTREAM project. Based on the Hamming distance model, the proposed attack exploits the resynchronization phase of Trivium. By choosing proper initial value vectors, the algorithmic noise of the device is completely eliminated. Furthermore, a novel concept of modified correlation coefficients is introduced, which can be used to describe the relation between the hypothetical power consumption values and the measured power consumption values. Through the calculation of modified correlation coefficients, the effect of the electronic noise is significantly decreased and values of the hypotheses can be discriminated uniquely by the highest modified correlation coefficient. According to the recovered hypotheses, many equations on the secret key bits can be obtained, which will be sequentially solved to extract the secret key of Trivium. Compared with Fischer's differential power analysis attack on Trivium, the proposed algorithm is more efficient and robust. Finally, a simulation attack is mounted to confirm the efficiency of the algorithm. Copyright © 2011 John Wiley & Sons, Ltd.