In this paper, decision-making models for orthodontic tooth extraction in treatment planning that are fully compatible with the digital associative processor has been developed. The architecture of the models was designed according to the specification of the digital associative processor. Feature variables were extracted from the pre-treatment clinical records of orthodontic patients and projected in the feature vector space by the 8-bit nonlinear transformation functions newly developed based on the expertise knowledge. Additionally, the fiducial treatments were defined by actual treatments described in the medical chart and judgments of three orthodontists having more than eight years of clinical experiences for each case. The sets of feature vectors and their corresponding fiducial treatments were employed as templates in the models. The N-neighboring search in the model templates was performed using weighed Manhattan distance as a dissimilarity measure to predict the optimum treatments for an input case. The hardware-friendly decision-making models for orthodontic tooth extraction were successfully developed and it was found that they were applicable in a clinical use.