In this paper, we outline a way to improve computer-assisted neurosurgery using surgical models along with patient-specific models built from multimodal images. We propose a methodological framework for surgical models that include the definition of a surgical ontology, the development of software for describing surgical procedures based on this ontology and the analysis of these descriptions to generate knowledge about surgical practice. Knowledge generation is illustrated by two studies. One hundred fifty-nine patients who underwent brain tumor surgery were described from postoperative reports using the surgical ontology. First, from a subset of 106 surgical cases, we computed a decision tree using a prediction approach that gave probability in terms of operating room patient positioning percentages and according to tumor location within one or more lobes. Second, from the whole set of 159 surgical cases, we identified 6 clusters describing families of cases according to pathology-related parameters. Results from both studies showed possible prediction of parts of the surgical procedure from pathology-related characteristics of the patient. Surgical models enable surgical knowledge to be made explicit, facilitating the surgical decision-making process and surgical planning and improving the human–computer interface during surgery.