Abstract: Tooth segmentation is the foundation of digital dental orthodontics.To solve the problem of low automatization and imprecision of existing algorithms,this paper proposed a semi-automatic tooth segmentation algorithm.This algorithm includes three steps:First,the feature point between every two teeth was marked.A concavity-sensitive harmonic field matrix was built on local tooth model using the two feature points beside the tooth,and the scalar field was obtained on the local mesh.Second,a series of isolines were obtained using the scalar field.The best isoline was selected using a voting mechanism.Third,a heuristic feature line extraction algorithm was used to optimize the selected isoline to obtain the precise gingival line,and realized accurate segmentation of teeth.All the dental models come from the clinic.Experiments show that our method is of strong compatibility,high accuracy and speed,and can meet practical requirements.