This paper proposes a method of quantitative evaluation of fault diagnosability based on maximum mean discrepancy(MMD). The method evaluates the fault diagnosability quantitatively by measuring the distance between measurement data under different fault conditions without building any system model. It is suitable for the system with complex structures that are difficult to build models and obtain measurement data. Firstly, the measurement data is mapped to the reproducing kernel Hilbert space(RKHS) through the characteristic kernel. The MMD is taken as the distance measure of multivariate distributions, and the fault diagnosability quantitative evaluation is transformed into the distance measurement of multivariate distributions in the RKHS. Then, the influence of measurement noise intensity on the result of fault diagnosability evaluation is analyzed by mathematical derivation. Finally, the validity of the proposed method is verified by simulation analysis.