During the Central Equatorial Pacific Experiment, ice crystal sizes and shapes were measured in an outflow anvil. A habit (i.e., column, bullet rosette, Koch fractal polycrystal, sphere) was assigned to each particle using a self-organized neural network based on simulations of how the maximum particle dimension and area ratio varied for random orientations of these crystals. Average ice crystal size and shape distributions were calculated for 25 km long segments at six altitudes using measurements from a two-dimensional cloud probe for crystals larger than 90 μm and a parameterization for smaller crystals based on measurements from the Video Ice Particle Sampler (VIPS). Mean-scattering properties were determined by weighting the size and shape dependent single-scattering properties computed with ray-tracing algorithms according to scattering cross-section. Reflectances at 0.664, 0.875, 1.621, and 2.142 μm were then calculated using a Monte Carlo radiative transfer routine. Although these reflectances agree reasonably with those measured by the MODIS airborne simulator (MAS) above the anvil, uncertainties in cloud base and system evolution prevent a determination of whether ray-tracing or anomalous diffraction theory better predict reflectance. The calculated reflectances are as sensitive to the numbers and shapes of crystals smaller than 90 μm as to those of larger crystals. The calculated reflectances were insensitive to the classification scheme (i.e., neural network, discriminator analysis, and previously used classification scheme) for assigning particle shape to observed crystals. However, the reflectances significantly depended on assumed particle shape. ¿ 1999 American Geophysical Union