An improved rule-based backward fuzzy reasoning algorithm was pointed. The algorithm is based on the fuzzy Petri net （FPN） which represents fuzzy production rules. The algorithm is also suit for a class of rule-based systems

i.e. it can deal with the fuzzy Petri net model which results from the real system. Via using the algorithm

one can calculate the tokens of any appointed places which correspond to the true fuzzy values of the relevant propositions. Through practical applications

the advantages of the proposed algorithm compared with those present results obtained by the existed algo- rithm.