Abstract: To address current limitations in sediment-type mapping and to achieve unbiased geostatistical inference of sediment composition, we developed and validated a standardized “Composition-Classification” mapping workflow based on grain-size composition data from about 3200 surface sediment samples taken from the Yangtze River Estuary–East China Sea shelf. The sand-silt-clay percentage data were transformed first using the additive log-ratio (ALR) transformation, followed by empirical Bayesian kriging (EBK) interpolation in the transformed space. An unbiased Monte Carlo (MC) back-transformation was then applied to recover the continuous spatial distribution of each component. At last, Folk’s classification scheme was used on a cell-by-cell basis to produce posterior mean-probability and maximum-probability sediment-type maps. The workflow simultaneously outputs several diagnostic measures of uncertainty, including maximum class probability, probability difference (ΔP) and normalized entropy (H_n). Results show that the method could inherently satisfy non-negativity and closure constraints, and yield sediment-type maps with smooth spatial transitions and highly consistent spatial patterns in the diagnostic indices. Meanwhile, it robustly delineated high-confidence areas while identified low-confidence zones that require further optimization. The proposed method also supported rapid remapping to different management units, providing a technical support for marine resource exploration and environmental management.