OBJECTIVE To develop and evaluate a logistics regression diagnostic model based on computer tomography (CT) features to differentiate tuberculous spondylitis (TS) from pyogenic spondylitis (PS). METHODS Demographic and clinical features were collected from the Electronic Medical Record System. Data of bony changes seen on CT images were compared between the PS (n = 61) and TS (n = 51) groups using the chi-squared test or t test. Based on features that were identified to be significant, a diagnostic model was developed from a derivation set (two thirds) and evaluated in a validation set (one third). The sensitivity, specificity, and area under the receiver operating characteristic curve (AUC) were calculated. RESULTS The width of bone formation around the vertebra and sequestrum was greater in the TS group. There were significant differences between the two groups in the horizontal and longitudinal location of erosion and the morphology of axial bone destruction and sagittal residual vertebra. Kyphotic deformity and overlapping vertebrae were more common in the TS group. A diagnostic model that included eight predictors was developed and simplified to include the following six predictors: width of the bone formation surrounding the vertebra, longitudinal location, axial-specific erosive morphology, specific morphology of the residual vertebra, kyphotic deformity, and overlapping vertebrae. The simplified model showed good sensitivity, specificity, and total accuracy (85.59%, 87.80%, and 86.50%, respectively); the AUC was 0.95, indicating good clinical predictive ability. CONCLUSIONS A diagnostic model based on bone destruction and formation seen on CT images can facilitate clinical differentiation of TS from PS. CONCLUSIONS • We have developed and validated a simple diagnostic model based on bone destruction and formation observed on CT images that can differentiate tuberculous spondylitis from pyogenic spondylitis. • The model includes six predictors: width of the bone formation surrounding the vertebra, longitudinal location, axial-specific erosive morphology, specific morphology of the residual vertebra, kyphotic deformity, and overlapping vertebrae. • The simplified model has good sensitivity, specificity, and total accuracy with a high AUC, indicating excellent predictive ability.