Soil bulk density, which can be measured through several labor-intensive procedures, is often missing from most soil databases. However, it is an essential parameter in calculation in many cases and models, and it is feasible to derive soil bulk density from some other attributes of the soil. As China has a huge variety of soil types, whether the existing pedotransfer functions (PTFs) are still applicable to the various soils calls for in-depth analysis. The soil data involved in this study were cited from the Second National Soil Survey of China, from the database for the Chinese Soil Taxonomy and from other publications, covering all the major types of soils in China. The data gathered during the Second National Soil Survey were converted into the Chinese Soil Taxonomy system by means of approximate reference between the two systems using a WebGIS-based inquiry system. After screening of the data in quality and depleting some abnormal values, a total of 2 441 complete soil datasets were obtained, covering all the major types of soils in China. In this paper, 2 published PTFs were evaluated and compared in prediction accuracy and applicability for different types of soils, and a PTF, the most adaptable to the major types of soils in China, was developed through regression analysis using SPSS. In addition, exploratory stepwise regression models were proposed and the parameters of the nonlinear model that used only the OM variable were revised based on taxonomical partitioning of the data. Results show that the two existing models varied in performance and were not high in prediction accuracy when used for some types of soils. The bulk density MPE values acquired with the polynomial model of Histosols, Halosols, Gleyosols, Isohumosols and Cambosols were negative (-0.06～-0.01), while those of Anthrosols, Ferralosols, Aridosols, Ferrosols, Argosols, Primosols and Vertosols were positive (0.00～0.02), indicating that the model overestimated BDs of the soils in the former group, and underestimated those in the latter group. The MPE values obtained with the nonlinear model of Aridosols, Halosols, Isohumosols and Ferrosols were negative (-0.02～-0.01), while those of Histosols, Anthrosols, Ferralosols, Gleyosols, Argosols, Cambosols, Primosols and Vertosols were positive (0.00～0.10), indicating the model overestimated BDs of the soils in the former group, and underestimated those of the soils in the latter group. Comprehensive comparison of the scattergraphs of MPE, RMSPE, R2 and measured and predicted BDs indicates that the two models did not vary much in prediction accuracy when used for soils of the same soil order. The PTF developed after the data of the soil taxonomy were partitioned significantly improved the model’s performance. The new PTF was pretty high in prediction accuracy when used for Histosols, Ferralosols, Gleyosols, Isohumosols, Ferrosols, Argosols, Cambosols, Primosols and Vertosols, but it tended to be low in accuracy when used for Anthrosols, Halosols and Aridosols. So the model should be used with care. Besides, in using PTFs, it is essential to pay special attention to area and scope they are applied to.