The work aims to study the influence laws of random errors on the prediction results by analyzing the transmission of random error in the inverse-substructuring analysis method based on the experimental frequency response functions (FRFs), so as to provide reference value for analysis of dynamic characteristics of complicated structures based on inverse-substructuring method. Different degrees of random errors were applied to the obtained system frequency response function, and the influence of each coupled system frequency response function on the predicted substructure frequency response function was compared and analyzed. After random error was applied to the coupled system frequency response function, the predicted substructure frequency response function after decoupling of the coupled system by the inverse-substructuring method seriously deviated from the true value, especially near the resonance frequency, and the applied random error was even magnified dozens of times in the predicted substructure frequency response function, resulting in unreliable prediction results. The frequency response function at the coupling point of the coupled system had the greatest influence on the prediction results. Through the analysis, it is clear that the random errors carried by the system frequency response function have an influence on the prediction results, and these errors will be amplified along with the matrix inversion operation, and the cross-coupled system frequency response function has the most significant influence on the prediction results.