During precision sowing, the contact process between the soil and seeds cannot be ignored. The constitutive relationship of soil is relatively complex, with characteristics such as high nonlinearity, while the contact mechanism between the soil and seeds is unclear. To better understand the contact between seeds and soil, it is necessary to establish a reasonable contact model. Ellipsoidal seeds, such as soybean, red bean, and kidney bean seeds, were adopted as research objects. In this paper, we used the discrete element method to establish an ellipsoidal seed–soil contact model. The JKR + bonding model was adopted for describing the adhesion between soil particles, and the Hertz–Mindlin new restitution (HMNS) model was used for ellipsoidal seed particles to eliminate the multiple contact point issue when modeling with the multi-sphere filling method. Moreover, both simulations and experiments were conducted to calibrate the interaction parameters between soil and seeds. The path of steepest ascent test and Box‒Behnken design (BBD) tests were also used, as well as direct shear tests. Thus, certain soil parameter values were obtained, namely the JKR surface energy was 4.436 J/m2, the normal stiffness per unit area was 2.86 × 106 N/m3, the shear stiffness per unit area was 5.54 × 105 N/m3, the critical normal stress was 1833 Pa, and the critical shear stress was 3332 Pa. In addition, the simulation parameters for ellipsoidal seeds were obtained from previous works. Moreover, to obtain more accurate ellipsoidal seed–soil interaction parameters, collision tests, static friction tests, and rolling friction tests were adopted. A single-factor test was used to calibrate the ellipsoidal seed–soil interaction parameters. The calibration results were as follows: the collision restitution coefficients of ellipsoidal seeds with soil were all 0.25. The static friction coefficient of soybeans with soil was 0.6, that of red beans with soil was 0.65, and that of kidney beans with soil was 0.5. The rolling friction coefficient of soybeans with soil was 0.1, that of red beans with soil was 0.14, and that of kidney beans with soil was 0.14. Finally, the rationality of parameter selection was verified through piling tests between ellipsoidal seeds and soil. The relative error of the angle of repose of soybean/soil was 2.99%, that of red bean/soil was 0.60%, and that of kidney bean/soil was 0.55%. Thus, the feasibility and rationality of the contact models between the ellipsoidal seeds and soil established in this paper, as well as the parameter selection, were verified.