Deciphering significant sedimentological processes froma set of sediment samples is an important step in reconstructing environmental changes. One approach going beyond classical methods is the unmixing of grain-size distributions. This paper presents a flexible end-member modelling algorithm that is based on eigenspace analysis and considers inherent uncertainties. It has been applied to the detrital grain-size components of lacustrine surface sediment samples of Lake Donggi Cona, Qinghai Province, China. It allows up to five grain-size endmembers to be characterised and quantified in an optimal model. An end-member with a major mode in the clay domain accounts for 34% of variance within the grain-size data set. It may represent sedimentation of suspension load from linear and laminar runoff during heavy precipitation events in summer. Three endmembers in the fine sand to medium silt domainsmake up 60% of lacustrine sedimentation. They may represent local to remote aeolian processes that peak in wintertime. A multimodal end-member explaining the remaining 6% of variance may represent further fluvial and littoral dynamics or random fluctuations and measurement errors. Several model runs of different scaling and numbers of end-members provided a suitableway to determine uncertainties inherent to the model. A comparison of 12 different model runs and their respective uncertainties yielded a distinct model of robust end-members. The clay and medium silt end-members are robust features of detrital sedimentationwithin Lake Donggi Cona. They alone explain 54.4% of total variance in the data. However, no spatial pattern or relation to water depth is found for any of the grain-size end-members. Thus,when past detrital sedimentation at Lake Donggi Cona is reconstructed, a special focus should be on the robust features attributed to aeolian and suspension-related sedimentation processes as well as on effective sediment mixing processes impeding a distinct correlation between grain size and spatial attributes. Further applications of the end-member modelling algorithm to other depositional environments are encouraged to demonstrate its universal applicability.