The invention discloses an equivalent order reduction-based diagnosability evaluation method. According to the method, firstly, the influences of process and observation noises and the uncertainty areconsidered and a diagnosability window model of a complex dynamic system is obtained. Secondly, based on the recurrence relation in the system, a diagnosability dimension reduction window model of the system is obtained. The burden of data storage and calculation is greatly reduced, and the diagnosability calculation efficiency of the system is improved. Finally, the diagnosability quantitative evaluation indexes of the complex dynamic system are obtained. According to the invention, the influences of process and observation noises and the uncertainty on the correctness of diagnosability evaluation results are comprehensively considered. Meanwhile, the influence of a time window model on diagnosability evaluation results is also considered. As a result, the diagnosability design level ofthe system can be maximally improved during the design stage of the system. A guidance basis is provided for the configuration of the system. The design method of the system is optimized and the design cost of the system is reduced.