Fuzzy controllers can be made to adapt to changing process and environment dynamics. One methodology to adapt the initial knowledge base to changing operating conditions is presented. The membership functions associated with the process controller output are adjusted in response to future or past performance of the feedback control system. A linearized process model is identified on-line and is used to predict the future performance of the controller. This performance index is used to adapt the controller as long as the identified process model is statistically reliable. If the process model is considered unreliable then an index related to the past performance of the controller is used to make the adjustments.