Iris recognition using complex valued 2D Gabor filters received a great success in broad applications. Such method encoded the iris phase information which was extracted according to the complex valued 2D filters? response. In this paper, we theoretically analyzed the complex-valued 2D Gabor filters. Based on analysis, we concluded that the real part of complex-valued 2D Gabor filter would reduce the whole system performance because it is not perfectly band passed filter. An improved algorithm then proposed to extract the phase information of iris image. The new method has more advantages such as lower false rate, much less space of codes and shorter time of encoding and matching. In order to reduce the disturbance of eyelashes, eyelids, specular reflections, or other noise, a matching template is designed by dilating the positions of them. Experimental results show that the latter method has an encouraging performance.