Precise functional localization plays an important role in deep brain stimulation for treatment of Parkinson’s disease (PD) to target the lesion of stimulation and observe spike morphology of the neural circuit for evaluating effect of stimulation. This research described a functional localization method in the brain of a cynomolgus monkey based on spike patterns recognition at different locations through machine learning algorithm. Through K-means algorithm cluster, spikes were sorted into different clusters, and the template of sorting was considered as spike pattern of each cluster. Four different spike patterns in cortex, three different spike patterns in white matter, and two different spike patterns in striatum were found through sorting in normal monkey, which were seen as features of different locations in normal monkey. While spike patterns found in PD monkey were totally different, including four different spike patterns in cortex, four different spike patterns in white matter, and five different spike patterns in striatum, considered as features of different locations in PD monkey. Cubic support vector machine (SVM) was used to train spike pattern recognition model for functional localization with accuracy of 100% in normal monkey, and the evaluation of trained model demonstrated reasonably excellent recognition accuracy of 99.5%. Weighted K-nearest neighbor (KNN) showed a better performance of accuracy (94.5%) of spike pattern recognition for functional localization in PD monkey than cubic SVM. In evaluation testing of the trained weighted KNN model, the accuracy reached to 96.1%. The results revealed that functional localization based on spike patterns recognition using machine learning algorithm would be an important tool in precise targeting and evaluating outcome not only for Parkinson’s disease, but also for other major brain diseases. © 2019, Springer-Verlag GmbH Germany, part of Springer Nature.