A digital system architecture for a three-input one-output zero-order ANFIS (Adaptive Neuro-Fuzzy Inference System) is presented. The proposed architecture takes into account that the training process is done off-line in the MATLAB environment. The system is implemented as a nonlinear-function generator for test purposes. Post-place and route simulation results obtained on a Xilinx Spartan-3 XC3S200 FPGA are presented. Its correct operation is verified by the results obtained for two chosen functions. These results show that the system is capable of achieving a close approximation of any of the functions with a fast response time. © 2012 IEEE.