Once the input values are given, the active rules in a fuzzy inference execution have been determined. Based on the observation, our approach is to identify the active rules before fuzzy inference execution. To achieve this goal, our architecture provides the following two mechanisms: (1) a mechanism to ignore the non-active rules before fuzzy inference execution; and (2) a mechanism to arrange the active rules for fuzzy inference execution. The proposed architecture has been implemented using a 0.35μm cell library. Implementation data show that, if the number of active rules is only 4, the inference speed can achieve 23.755 MFLIPS. To the best of our knowledge, our approach is the fastest hardware implementation.