In this work, improved performance is obtained in human iris matching systems using a Fourier-based description to approximate non-circular pupil boundaries in human eyes. The study also leads an analysis of pupil shape. Excellent fitting is obtained for non-ideal cases such as oblong, irregular, off-centre and dilated pupils, and improved iris normalization is obtained compared to best fit circles. The method is applied to 1912 eye images of 478 eyes from the Bath database and the effect of increasing the number of Fourier coefficients on the pupil outline accuracy is studied. The RMS pixel error between the outline and actual edge points is seen to decrease from 1.48 (for circles) to 0.34 (for 9 coefficients). Only 21 cases are found to produce low deviations from a circular boundary, indicating that a majority of pupils (98.9%) are non-circular. The Equal Error Rate (EER) in verification using the new method is estimated to be significantly reduced, at 7.5?105 compared to 1.2?104 for pupils fitted as circular.