[Objective] Cation exchange capacity of clays is an important index for determining diagnostic horizons and diagnostic characteristics in the Chinese Soil Taxonomy (3rd edition) and the United States Soil Taxonomy. However, some studies have shown that the values of CEC₇ particles predicted using models are often higher than their corresponding measured ones, thus leading to misjudgment of taxon at high taxonomic levels. With the zonal soil in Jiangxi Province taken as the main object, this study aimed to optimize the current CEC prediction model to alleviate the impacts of its error factors, to build up a new model, based on main error factors, for predicting CEC₇ of clay particles, so as to improve prediction accuracy and to provide reliable data support for retrieval in the soil taxonomy.[Method] To that end, an idea of how to optimize the current model was put forward, suspicious error factors were screened out based on the previous researches and collated with those in the current model for correlation analysis, and error law in the current model was explored. Then soil samples were classified in line with the law to improve the model in prediction accuracy. Main error factors in each classification sample were searched out and got involved in modeling. Eventually, the optimized model for soil classification was established.[Result] By comparing the value estimated with the current model with the measured one, it is found that the former is generally higher than the latter. Previous studies have shown that soil organic matter, silt CEC₇, soil pH and soil free iron oxide content are factors affecting CEC of the fine soil in the B layer of weathered soil. Correlation analysis was performed of the factors with the error, and indicated that organic matter and silt CEC₇ were the main ones causing errors. Studies found that in predicting soils higher or lower than 6 g·kg^-1 in organic content, errors varied in dispersion. Therefore, in this study, all soil samples were sorted into two groups, high and low in organic matter content for modeling. For the group high in organic matter, errors were ultra-significantly related to soil organic matter content (R²=0.402, n=23). Considering that organic matter may get bonded with clay particles, the group of samples high in organic matter content were further sorted in three subgroups, i.e. "clay soil samples", "clay loam soil samples", and "loam soil samples" and a model was set up for each of the three subgroups. In the subgroup low in organic matter content, errors were ultra-significantly related to CEC₇ of silt (R²=0.675, n=23), so a direct model was obtained for soil samples low in organic matter. Considering that CEC₇ is not easy to be measured, soil pH, annual mean temperature (Tem℃) and Latitude (Lat) were selected and used in modeling for predicting silt CEC₇, and consequently an indirect model based on environmental factors was established for soil samples low in organic matter. Through the accuracy evaluation of the models, it is found that optimization of the models has brought predicted values closer to measured values, and the models for all subgroups of soil samples are good in accuracy. Optimization of the models has raised retrieval of iron-rich soils from 20% to 93.3% in accuracy.[Conclusion] Based on the above findings, it is found that modeling by content of soil organic matter is reasonable. By analyzing sources of the errors with the current model and following the optimization formula, models for predicting cation exchange capacity of clay particles are established by content of soil organic matter with higher accuracy. The models may provide reliable data support for retrieval in the soil taxonomy.