The performance of iris recognition system is determined largely by the iris feature extraction algorithm. To improve the accuracy of iris recognition system, we propose an efficient algorithm for iris feature extraction based on 2D Haar wavelet. Firstly, the iris image is decomposed by the 2D Haar wavelet three times, and then a 375-bit iris code is obtained by quantizing all the high-frequency coefficients at third lever. Finally we use similarity degree function as matching scheme. Experimental results on CASIA iris database show that our algorithm has attractive performances than algorithm from Boles and Wildes, only inferior to Daugman’s algorithm,but the length of our iris code is only one-fifth comparing with Daugman’s. The proposed algorithm has the encouraging correct recognition rate (CRR) which is 99.16%, accompanying with very low equal error rate (EER) 0.54%.