Abstract: The planting area of garlic in China accounts for more than 90% of the whole area in the world. The major producer of garlic also needs mechanized harvesting in modern agriculture in recent years. Two kinds of harvesting technologies are divided mainly into: the widely-used segmented harvesting for the separation of garlic and soil individually, whereas, the new combined harvesting to concurrently separate the garlic, stem, root, and soil. However, the current garlic combine-harvester in the world can only collect the garlic under the upright position and wide rows, but cannot achieve the harvesting and cutting roots of lodging garlic plants. In this study, a novel garlic combined-harvester was designed to improve the harvesting efficiency, while reducing the labor intensity, according to the current state of garlic planting. Some specific conditions were also considered, including the working process of digging, the posture correction of garlic seedling, clamping, cutting, and low-damage collecting. The parameters of key components were determined using a theoretical and dynamic simulation of the operation process. A Box-Behnken orthogonal test was performed, where the forward speed, digging depth, and clamping distance were taken as the test factors, whereas, the damage rate and loss rate were the evaluation indices. A three-factor three-level Box-Behnken simulation was also carried out. A field test was conducted at a planting cooperative in Mindong, Shandong Province of China in May 2019. The test object was set as a variety of "Jinxiang Red Garlic". The shaft of shovel and gearbox gears were adjusted to change the distance between the clamping chains, thereby obtaining the different forward speeds, digging depths, and chain spacing during the test. The damage and loss rates were recorded for the harvested garlic in the field. Variance and response surface were utilized to determine the effects of forward speed, digging depth, and clamping distance on the evaluation index. A Design-Export software was used to optimize the model. An optimal combination of parameters was achieved as followed: the forward speed of 0.51 m/s, digging depth of 97.2 mm, and clamping distance of 7.6 mm, corresponding to the damage and loss rates of 0.63% and 1.25%, respectively. The parameters were tested in the field to verify the accuracy of the optimized model. The relative errors of all indices in the predicted and experimental data were less than 5%, indicating that the model was reliable and suitable for prediction and further optimization. The main factors affecting the damage and loss rates of garlic were determined for the forward speed, digging depth, and clamping distance. The finding can provide a sound theoretical reference for the design and optimization of garlic combine-harvester in mechanized production of intelligent agriculture.