Cirrus clouds are vertically inhomogeneous as evidenced by numerous in-situ samplings and lidar/radar observations. We explored the possibility of retrieving the vertical profile of cirrus cloud particle sizes using MODIS 0.645, 1.64 and 2.13 and 3.75 μm channel reflectance/radiance, and developed a conceptual approach for inferring the linearly constrained vertical profile of cirrus cloud effective diameter as a function of cloud optical depth. This approach uses a successive minimization method to search for the set of cloud parameters that is associated with the minimal difference between simulated and observed reflectance/radiance. The simulated reflectances/radiances are computed using an adding-doubling radiative transfer model, and are subject to correction parameterizations for multiple-reflection between surface and cloud, 3.75 μm thermal infrared emission, and Rayleigh scattering. We investigated the accuracy of this approach by synthetic retrievals subject to prescribed instrument noises and uncertainties in input parameters, including surface albedo at each of the four MODIS channels, cloud top pressure, cloud top temperature and surface temperature. The retrieval algorithm has been applied to MODIS data for two selected cirrus scenes, one over the DOE-ARM-SGP surface observation site and the other over southern Texas coastal area, and the retrieved vertical profiles of effective diameter are compared with those derived from MODIS cloud product, collocated and coincident ground based Millimeter-wave Cloud Radar or aircraft in-situ cloud particle imager observations. Our retrieval results correlate well with these independent measurements.