Objectives: Single isocenter multiple target (SIMT) radiosurgery treatment with volume modulated arc therapy (VMAT) reduces treatment time significantly compared to single isocenter per target treatment. However, the treatment planning process is complicated and time consuming, and the plan quality is highly dependent on planner's experience and knowledge. In this study, a knowledge-based optimization model was created from previously treated plans and implemented for plan quality control. Methods: Fifty SIMT VMAT treatment plans were selected to train the knowledge-based optimization model using RapidPlan (Eclipse V13.6, Varian Medical Systems). These training plans were prescribed with either one fraction (12 - 22 Gy/fx) or five fractions (5.5 Gy/fx) with 1 - 24 targets of volume ranging from < 0.1 to 116.5 cm3. Typical beam arrangements consisted of 4 - 5 partial arcs spanning 180 degrees using 6FFF energy and couch positions of 0, 45, 90 and 315 degrees. All plans were normalized to 100% prescription dose covering at least 99.5% of all planning target volumes (PTVs). The optimization model was then validated and used to assess another twenty previously treated cases. Three types of plans were created (Automatic Plan, User Plan, and Clinical Plan) for comparison. Automatic Plans started with model-predicted DVHs and went through one round of optimization without any manual adjustment. User Plans started with userdefined objectives/priorities and also went through one round of optimization. A planner then further optimized each User Plan to get the Clinical Plan. Dosimetric parameters including maximum dose (Dmax) to the organs at risk (OARs), PTV Dmax, PTV coverage, conformity index (CI), and monitor units (MU) were compared for plan evaluation. Results: In general, knowledge-based Automatic Plans were comparable to Clinical Plans and superior to User Plans. The PTV coverage and CI in Automatic Plans were equivalent to those in Clinical Plans. Dmax for brainstem, optical chiasm, optical nerves, eyes, and lenses in Automatic Plans were lower or equivalent to those in Clinical Plans, while PTV Dmax was higher for all Automatic Plans. As a quality assurance tool, the knowledgebased model also identified two Clinical Plans with inferior plan quality. For the first identified case, all dosimetric parameters were clinically equivalent or better for Automatic Plan compared to Clinical Plan. At the meantime, the total MUs for Automatic Plan were reduced to 1477 compared to 2311 for the Clinical Plan (a 36% reduction). For the second identified case, Dmax to left optical nerve was 30% lower for the Automatic Plan, a significant reduction of 8.3 Gy. Knowledge-based Automatic Plans indicated that plan quality for both Clinical Plans could have been further improved. Conclusions: A knowledge-based model was created for SIMT planning with VMAT technology. Results indicated that Automatic Plans generated based on the model easily met the institutional or RTOG objectives for OARs, and were comparable to manual plans in key DVH parameters for both PTV and OARs. In addition, the model based SIMT optimization process was independent of planner's knowledge and experience and substantially reduced optimization efforts since no human interaction was involved during the optimization. This knowledge-based model can be used to provide guidance for SIMT VMAT planning and also serve as a quality assurance tool to evaluate clinically treated plans.