Drilling and blasting is still the most efficient way to explore deep phosphate mine excavation and mining. There is a severe constraint on the efficiency of phosphate mine digging as its level remained at 70 to 80 meters every month for many years. Therefore, the ore rock blastability classification is critical for the deep phosphate mine working face. The longitudinal wave velocity tests of the rock body in an underground phosphate mine in Yichang, Hubei Province, and measurements of physical and mechanical properties such as rock density, uniaxial compressive strength and tensile strength were carried out. The rock density, uniaxial compressive strength, tensile strength, and rock integrity coefficient were obtained for four types of rocks, namely, dolomitic striped phosphorite, dense striped phosphorite, argillaceous striped phosphorite, and carbon-bearing argillaceous dolomite. To complete the deep phosphorite workings of the mine rock blastability classification, a BP neural network model was established by stochastic functions to generate a large number of learning and testing samples using the Matlab neural network toolbox as taking the pre-measured rock density, uniaxial compressive strength, tensile strength and rock integrity coefficients as inputs and the rock blastability classification as outputs. The grading results show that dolomite-banded phosphorite and mud-banded phosphorite are moderately blastable, and dense-banded phosphorite and carbonaceous mud dolomite are difficult to blast. According to the classification results, the blasting parameters of the stope can be optimized to enhance the blasting effect, reduce the single consumption and the bulk rate of explosives, and improve the safety and economic benefits of deep phosphate mining.