Accurate　assessment　of　regional　soil　bulk　density　is　a　key　process　to　understand　the　role　of　soil　organic　carbon　in　the　global　carbon　cycle.　Using　the　second　national　soil　survey　data　of　China,　this　study　applied　three　machine　learning　methods:　random　forest　model,　artificial　neural　network,　and　support　vector　machine　model,　to　estimate　the　soil　bulk　density.　The　results　show　that　the　correlation　between　soil　bulk　density　and　soil　organic　matter　is　the　best,　and　its　correlation　coefficient　is　greater　than　0.5.　Among　the　three　estimation　models,　the　random　forest　model　has　the　highest　prediction　accuracy,　with　a　coefficient　of　determination　R2　of　0.　53,　RMSE　of　0.　1471g/cm3,　and　SDPE　of　0.1475g/cm3.　In　order　to　reduce　the　contingency　of　data　division,　a　10-fold　cross-validation　was　performed　on　each　model.　The　results　showed　that　the　RMSE　and　SDPE　of　the　random　forest　model　were　the　smallest,　and　the　overall　trend　of　MPE　was　closest　to　0,　and　the　model　fitting　effect　was　good.　According　to　the　soil　data　of　this　study,　using　the　soil　pedotransfer　functions　(PTFs)　published　in　China　to　predict　the　soil　bulk　density,　in　the　estimation　results,　the　prediction　accuracy　and　model　fitting　effect　are　not　as　good　as　machine　learning　methods.　The　estimation　of　soil　bulk　density　based　on　machine　learning　methods　is　generally　close　to　the　actual　value,　indicating　that　machine　learning　methods　can　be　applied　to　soil　bulk　density　estimation.　©　2022　IEEE.