Abstarct: Many previously lost opportunities for individuals who have suffered from different forms of amputation or were, in fact, born with a congenital physical impairment, such as the ability to work or pursue a chosen way of life, can be somewhat restored using artificial prostheses. In addition, the electromyo control system can be considered as the major component in a wide range of modern prostheses, this system utilizes Electromyogram (EMG) signals, stemmed from the individual’s muscles in order to control and also change the position of the prosthesis. Although numerous research has been conducted in this field involving the detection and the restoration of functional hand movement, few studies have indeed investigated this concept in the detection of single and combined movements of fingers. As a result, the purpose of the study in this dissertation was to develop an optimal approach for Electromyo signals classification of various single and combined movements of fingers. A considerable part of this thesis therefore meticulously examines the detection of single and combined movements of fingers using surface EMG signals, such that the positions of robotic hand’s fingers can be fully controlled in response to these signals. In addition, the number of electrodes play a crucial role in the signal accumulation and also the classification accuracy. In other words, the more electrodes used, the higher the movement classification accuracy, while on the other hand, operating costs can evidently increase. Thus, we made a conscious effort to achieve the highest possible accuracy while utilizing the minimum number of electrodes. Therefore, two EMG electrodes, located on the forearm, were employed to collect EMG data from ten participants. In order to ensure maximum detection capability with respect to the different finger movements, a collection of various features in two domains of time and time-frequency was extracted. It should be noted that Stockwell transform was incorporated so as to analyze the time-frequency domain. In order to diminish the computational costs and unnecessary dimensions in the analysis, Principal Component Analysis (PCA) algorithm was utilized to reduce the number of features such that the classification accuracy level remains rather constant. Finally, three algorithms of Artificial Neural Network (ANN), Support Vector Machines (SVM) and k-Nearest Neighbors (KNN) were implemented to classify the obtained data and also determine the movement classifications. Practical results and statistical tests revealed that the suggested method, with an average accuracy of %92/28±0/55, was, in fact, able to adequately classify various finger movements. Moreover, compared to other studies, more than satisfactory results were achieved which essentially show the significance of our proposed method.