Mesoscale models that predict the temporal evolution of tropical cyclones (TCs) are sensitive to the representation of cloud microphysical processes through their effect on modeled latent heat release. The cloud parameterizations used in such models make assumptions about the size distributions (SDs) of different ice species, such as cloud ice, graupel, and snow, which have typically not been based on observations obtained in TCs. The representativeness of these parameterizations for TCs is not well known.In this study, in-situ observations of ice hydrometeor SDs acquired in tropical storms, depressions and waves during the NASA African Monsoon Multidisciplinary Analyses project are fit to gamma functions to determine how the intercept (No), shape (''''), and slope ('''') parameters vary with environmental conditions such as the total water content (TWC), vertical velocity (w), temperature (T), and TC stage of development. The No, '''', '''' solution representing the best fit is determined by forcing three moments of the fit distributions to match as closely as possible the corresponding moments computed from the observed SDs that are truncated between some minimum and maximum dimension. This is done using a non-linear Levenberg-Marquardt algorithm that minimizes a c2 difference between the fit and observed distributions. A surface of equally plausible solutions in No-''''-'''' phase space is defined as all solutions whose c2 difference is within some small ''''c2 of the minimum c2 for each SD. Families of SDs were determined for different environmental conditions (e.g. SDs found for updrafts, downdrafts and stratiform regions for w). A range of No, '''', and '''' that characterize the different environmental condition was determined. To test the significance of the variation of SDs for different environmental conditions, frequency distributions of mass-weighted terminal velocities (Vq) were calculated. The relative importance of uncertainties in fits to individual SDs and of dependence on environmental conditions in controlling the variability of No, '''', and '''' and Vq is discussed.