Mineralogical information obtained from plant surveys is subject to errors, both due to sampling and experiment. To reduce these errors, and to ensure that information is consistent, a data reconciliation model has been developed for mineralogical data. Its capability includes mass balancing size data, total assays, assays within size-classes and liberation data. Unlike any other mass balance package, this package also allows reconciliation of the complex mineralogical information. For example, particles cannot simply be treated as binary, but consist of combinations of many minerals. Hence the distribution of mineralogical information is multidimensional and the mineralogical distributions are referred to as multiphase. In this paper, the different levels of mass balancing are explained, including how to mass balance liberation data. Furthermore a new technique based on probability theory can be used to infer distributions even when there is insufficient information to perform a mass balance, and this technique is also explained. Results for various stages are given.