As an important member in biometrics technology, iris recognition can be performed widely in the application future and has a wide market development future, particularly in the security for certain place or resources. In this paper, a new algorithm for recognizing irises is proposed, which utilizes Kernel principle component analysis (KPCA) to extract iris texture feature and adopts competitive learning mechanism to recognize. Experiment results show that the algorithm is efficient and adaptable to environment. Therefore, it is of significant value in application.