This paper presents a novel scheme for cancelable iris recognition based on comb filtering. This scheme begins with a coarse-to-fine iris localization stage. After that, Gabor filtering is applied for feature extraction. The two-dimensional phase pattern of features generated with the LogGabor filter is distorted through comb filtering. The objective of this distortion process is to generate a cancelable feature pattern that represents the iris. The ability to reinitiate a new cancelable pattern is guaranteed through the variation of the comb filter order. The proposed scheme is compared with a cancelable random projection scheme for iris recognition. Experimental results are conducted on CASIA-IrisV3-Interval database for both random projection and comb filtering schemes. Moreover, evaluation metrics are estimated for different comb filter orders of 6, 8, 10, and 12 in addition to the case of original iris features. Hamming distance and Receiver Operating Characteristic (ROC) curve are estimated for both random projection and comb filtering schemes to check robustness and stability. The experimental results show a significant gain in both privacy and performance. Also, the comb filtering scheme achieves a superior performance for all orders compared to the random projection scheme. The proposed comb filtering scheme achieves the highest accuracy of 99.75% for order 6 and a promising Equal Error Rate (EER) of 0.36% for order 10.