IrisCode, a widely deployed iris recognition algorithm, developed in 1993 and continuously modified by Daugman has attracted considerable attentions. IrisCode using a coarse phase representation has number of properties such as rapid matching, binomial imposter distribution and predictable false acceptance rate. Although many similar coding methods have been developed for irises and palmprints based on IrisCode, a theoretical analysis of IrisCode has not been provided. In this paper, we aim at studying (1) the nature of IrisCode, (2) the property of the phase of Gabor function, (3) the extension of bitwise hamming distance and (4) the theoretical foundation of the binomial imposter distribution and extending the coarse phase representation to a precise phase representation. Precisely, we demonstrate that IrisCode is a clustering algorithm with four prototypes; the locus of a Gabor function is a two-dimensional ellipse with respect to the phase parameter and bitwise hamming can be regarded as angular distance. Using these properties, we provide a precise phase representation for IrisCode with an effective implementation for filtering and matching. Practically, the imposter distribution of IrisCode follows binomial distribution. However, the theoretical evidence is incomplete according to our analysis.