This research has focused on the system identification of a two -link manipulator.In order to identify more accurately the system under consideration and comprehend the system complexity, the manipulator identification was completed in four stages, stage 1 and 2 correspond to identification of link 1 and 2 as two different single-link manipulators.IN satge 3, the complete system, i.e. the two-link manipulator, was identified. In the final stage, the identified model of the system was used to control either single-link manipulator, partically. Due to thedifferent behavior of the system in high and low veliocity regimes, al the identification process were repeated for these two cases. thus, a switching model was developed to consider both regimes. The identification process started with collecting motors' voltage and joint's position as the input and output data of the manipulators. Although a time domain approach has been taken into account in the system identification, filtering and differentiation of the joints' position was carried out in the frequency domain, using a cutting window in the FFT of the data. The joints'velocity and accelerationdata were obtained with multiplying the filtered position data byjw and jw^2, respectively.In the nonlinear greybox structure utilized in this research, the presented model considers dynamics of the DC motors, flexiblity, friction and backlash of the joints. Friction and backlash are modeled by newly introduced aanti-symmetric nonlinear models. A two-mass flexible model is used to model the joints' flexibility. Following the optimization process for test data and proposed model, a statistical approach has been implemented to assess the model as well as to select the most appropriate parameters obtained from the the optimization process.In the final stage of the research, the identified model and parameters was utilized to design and implement controllers for both single-link and two-link manipulators. The controllers were designed using a CTM approach. For the sake of comparison three different models of the sysem were used to design the controllers. The first one is a mathematical model with calculated parameters from design softwares. The second model is similar to mathematical model; however, instead of the calculated parameters we have used the parameters that identified in this research. For the last case, the identified model and its parameters were used. Comparison of the experiment results showed a great improvment (about 95%) in the closed loop system performance from the mathematical model based controller to identified model based controller.