A m ethod ofdesigning an adaptive fuzzy predictive controlsystem is pro- posed. The controlsystem includes a fuzzy predictor and a fuzzy predictive controller, w hich are trained by BPalgorithm ,added new fuzzy rules according to the firing levelofeach ofthe rules(in the rulebase), and reduced thecontrary rulesto m eetthecom pletenessoffuzzy rules. The controlsystem hasstillthegood perform ance fornonlinearplantw ith dead tim e even after the training is stopped.