In this thesis, a new generic CMOS circuit for realizing different nonlinear functions from the same topology is presented. This circuit can be very useful in Neural Networks applications as it implements the main nonlinearities required by many types of Neural Networks. With transistors operating in strong inversion the circuit can be digitally configured to realize any of the following four functions: Gaussian (Radial Basis), Sigmoid and two piecewise linear functions – Triangular and Satlin. The circuit can approximate these functions with RRMS error less than 1%. It is shown that the center, width, peak amplitude and slope of the dc transfer curve can be independently controlled. Simulation results using 0.18 CMOS process model parameters of TSMC technology are included. Keywords: Analog Circuits, Programmable Neural Networks.