The stiction of control valve is one of main causes which make control loop instable in the process industry. Now a number of studies on the diagnosis of control valves are based on stiction. It needs developing a proper method which can detect, quantify, and compensate stiction. A non-invasive method has been used into the fault diagnosis for stiction of pneumatic control valve. Firstly, a two-parameter data-driven model about stiction has been established, and on the basis of separable least squares method, combined with the global search algorithm to deduce the model identification method. Secondly, ellipse fitting method has been adopt to confirm whether the stiction existed, thus the stiction could be quantified, and on this basis, the stiction was compensated. So it improved the performance of control system. Thirdly, the validity of the method was confirmed by simulation.