Objective: The aim was to assess the intraobserver and interobserver reliabilities of temporomandibular joint linear measurements and condylar shape classifications performed with cone-beam computed tomography (CBCT). Methods: CBCT images of 30 patients were measured at two different time points by two orthodontists using the Dolphin 3D program (n = 60). Anterior, posterior, and superior joint space measurements and sagittal joint morphology classification in the sagittal view and medial and lateral joint space and mediolateral width measurements and coronal joint morphology classification in the coronal view were recorded. Intraclass-interclass correlation coefficients (ICC) and kappa statistics were used to assess intraobserver and interobserver reliability for the measurements and morphology classifications, respectively. Results: The ICC values were good for measurements of the posterior joint space by observer I and for measurements of the posterior, medial, and lateral joint spaces by observer II, while the other intraobserver measurements were excellent. Only the mediolateral width measurements showed excellent interobserver ICC values, while the other measurements showed good interobserver ICC values. Intraobserver agreement for the sagittal morphology classifications was moderate (kappa = 0.479) and almost perfect (kappa = 0.858) for observers I and II, respectively, while the corresponding agreement for the coronal morphology classifications was substantial for both observers. The interobserver agreement values for sagittal and coronal morphology classifications were slight (kappa = 0.181) and fair (kappa = 0.265), respectively. Conclusions: Linear temporomandibular joint measurements were reproducible and reliable in both intraobserver and interobserver evaluations. However, interobserver agreement for assessments of condylar shape was low.