Aiming to meet the requirement of precision seeding operations in hill sloping fields and decrease the impacts of rightward or leftward inclination of metering devices on seed-filling quality, a self-suction mung bean precision seed metering device was designed. The structural parameters of spoon-shaped holes were determined through analysis of seed forces. The structural parameters of reciprocating suction devices were determined through calculating the vacuum degree of adsorbed single seeds. Kinematic simulation of the reciprocating suction device on the multi-body dynamics software ADAMS uncovered the temporal curves of the piston displacement and velocity. Then the simulation results were validated to be reliable. The seed-filling layer height was fixed at 70mm and the rotation speeds were fixed at 115r/min and 125r/min, while the right and left inclination angles were set at -12°, -6°, 0°, 6° or 12° (direction inclined to the seed chamber was negative). Comparison of three types of seed plates showed the spoon-shaped holes designed here facilitated the auxiliary seed filling under the left and right inclined status. Two-factor and five-level horizontal rotation combinations were designed and tested, involving factors of metering shaft rotation speed and filling layer height, and the indices of missing index, multiple index and qualified index. Then the test data were analyzed on Design-Expert. The missing index and qualified index were both more significantly affected by the metering shaft rotation speed than by the filling layer height. The multiple index was more considerably impacted by the filling layer height than by the metering shaft speed. At the metering shaft speed of 138r/min and the filling layer height of 65mm, the missing index, multiple index and qualified index were 2.97%, 3.43% and 93.58% respectively, which all met the national standards. The seed metering device can be referred during the design of precision seed metering devices in the future. Finally, the measured and predicted values of the regression model were obtained through the verification test. The relative error of the missing index was 4.4%, the relative error of the multiple index was 2.6%, and the relative error of the qualified index was 0.2%, which was basically consistent with the optimization result, proving the rationality of the regression model.