<p>Shrubs and small trees, which are essential elements of the understory vegetation, play a vital scientific role in accurately estimating biomass for assessing carbon storage in forest ecosystems. This research examined shrubs and small trees across 101 sample plots of broadleaf mixed forests in the Maoershan region. A total of 1,562 individual plants were destructively sampled to gather measured data. The Seemingly Unrelated Regression (SUR) method was utilized to create both genus-specific biomass SUR models (SUR-genus) and multi-genera generalized SUR models (SUR-total) for shrubs and trees separately. Additionally, by incorporating genus-level random effects into the SUR-total model, a Seemingly Unrelated Mixed-effects Model (SURM-total) was established. The reliability of the models was assessed using the jackknife method. The findings revealed that in the SUR-genus models, the most effective predictor variables were plant height (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\:H\)</EquationSource> </InlineEquation>) and crown area (<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\:CA\)</EquationSource> </InlineEquation>) for shrubs, while for trees, plant height (<InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(\:H\)</EquationSource> </InlineEquation>) and basal diameter (<InlineEquation ID="IEq4"> <EquationSource Format="TEX">\(\:BD\)</EquationSource> </InlineEquation>) were the key predictors. The adjusted coefficients of determination (<InlineEquation ID="IEq5"> <EquationSource Format="TEX">\(\:{R}_{a}^{2}\)</EquationSource> </InlineEquation>) varied from 0.705 to 0.977, with root mean square errors (RMSE) ranging between 0.095 and 14.307. In the SUR-total models, crown area and plant height were utilized for shrubs, whereas basal diameter and plant height were employed for trees to create binary biomass prediction models. The <InlineEquation ID="IEq6"> <EquationSource Format="TEX">\(\:{R}_{a}^{2}\)</EquationSource> </InlineEquation> values for these models exceeded 0.8. However, the SUR-total model exhibited significant systematic bias in estimating the biomass of different components for both shrubs and small trees, and its predictive performance was notably inferior to that of the SUR-genus models. The SURM-total model, which integrated genus-level random effects, effectively addressed interspecific differences and greatly enhanced prediction accuracy. In comparison to the SUR-total model, the <InlineEquation ID="IEq7"> <EquationSource Format="TEX">\(\:{R}_{a}^{2}\)</EquationSource> </InlineEquation> improved by 2.67% to 12.82%. The goodness-of-fit for the biomass of various shrub and small tree components was nearly identical to that of the SUR-genus models, with an average <InlineEquation ID="IEq8"> <EquationSource Format="TEX">\(\:{R}_{a}^{2}\)</EquationSource> </InlineEquation> difference of merely 0.01. A thorough analysis indicates that when estimating biomass for shrubs and small trees in broadleaf mixed forests, the genus-specific SUR-genus model is still the favored option. Nevertheless, the SURM-total model, known for its robust generalization ability, shows considerable promise for cross-species biomass estimation.</p>