The development of shrub allometric models is crucial for accurate biomass assessment, as well as for scientific studies of carbon storage and carbon cycling of desert ecosystems. The aim of the present study was to construct allometric models to predict biomass using easily measured variables for xerophytic shrubs. The 12 most widespread shrub species of northern China were selected and a total of 385 individuals were harvested to obtain the weight of its components (leaves, twigs, branches, and roots), the crown area (CA) and plant height (H). Based on a high coefficient of determination (R2), a low standard error of estimate (SEE), and low Akaike information criterion (AIC) values, 72 species-specific and 24 multispecies models with CA and H as independent variables were developed. The function lnW (biomass of different components) = a + b × lnX (predictor variable) was selected as optimal model. CA was revealed as the best independent variable for the biomass of leaves and twigs, and V (CA × H) was the best predictor variable for branches, aboveground, belowground, and total biomass. In conclusion, for the first time species-specific and multispecies models were constructed with a high goodness of fit of leaves, twigs, branches, aboveground, belowground, and total biomass for 12 shrub species in northern China. Compared to multispecies models, species-specific models had improved accuracy. Since biomass quantification is the basis of carbon stocks estimation, the models presented here can be considered as alternative tool for assessing carbon storage and carbon cycling of desert ecosystems.