Currently,　China　will　promote　county　towns＇　urbanization,　and　few　studies　have　analyzed　spatial　effects　of　urban　land-use　intensity　on　CO2　emissions　and　predicted　CO2　emissions　at　the　county　level　accurately.　Here,　taking　data　from　2010　to　2015　at　the　county　level　of　Zhejiang　Province　as　an　example,　we　analyzed　the　spatial　effect　of　urban　land-use　intensity　from　three　aspects　of　input,　density　and　output　on　CO2　emissions　by　the　spatial　Durbin　model　(SDM).　And　then　a　machine　learning　method　of　Back　Propagation　Neural　Network　(BPNN)　was　proposed　to　predict　CO2　emissions　for　2035　nonlinearly　under　the　different　promotions　of　urban　land-use　intensity.　The　main　result　and　conclusion　showed　that:　(1)　The　spillover　effects　of　urban　land-use　density　were　negatively　related　to　CO2　emissions,　and　urban　land-use　output　intensity　showed　positive　spillover　effects　on　CO2　emissions;　(2)　The　prediction　of　BPNN　showed　that　the　improvement　of　urban　land-use　intensity　would　not　be　effective　in　CO2　emissions　reductions　for　southwestern　counties　with　both　low　levels　of　urbanization　and　urban　land-use　in-tensity;　(3)　It　was　effective　in　CO2　emissions　reduction　by　slowing　down　the　growth　rate　of　urban　land-use　capital　input　intensity,　especially　for　the　northeast　region　which　was　developed　by　the　port　economy.　Our　study　encouraged　a　regional　differentiated　urban　land-use　intensity　improvement　strategy　for　Zhejiang　to　achieve　low　carbon　development.