This paper examines the sensitivity of retrieved ice particle sizes using split-window method to the light scattering program for the single scattering calculation. We find that for randomly oriented hexagonal ice particles the retrieval algorithm using the anomalous diffraction theory (ADT) significantly overestimates the mean effective ice particle sizes, D ge. The retrieved D ge based on the geometric optics method (GOM) and Mie theory agrees with reference results within ∼20% when D ge <30 μm . Based on the speculation that there is no “tunneling” for complex particles, some recent studies suggest that the ADT is an appropriate method to simulate the absorption coefficient for irregularly shaped particles in the infrared. In this study, however, we find that the overestimation of D ge due to the ADT is largely caused by the neglect of refraction and reflection processes, instead of by the neglect of “tunneling” in the absorption calculations. By considering complex particle shapes such as aggregates with surface roughness, we further find that the retrieved D ge based on the GOM is not sensitive to the particle shapes. Note that both ADT and GOM do not consider the “tunneling”, but the retrieved D ge based on the ADT is about two times larger than those based on the GOM. “Tunneling” plays a significant role in the retrieved D ge only when the D ge is larger than ∼30– 35 μm . In this study, we also examine the sensitivities of retrieved D ge to the ice particle size distributions assumed in the retrieval algorithm and to the errors in the emissivities. It is found that when the D ge is larger than ∼30– 40 μm , the retrieved D ge becomes very sensitive to the uncertainties related to the ice particle size distributions and to the errors in cirrus emissivities derived from measurements.