The invention discloses a model mismatching diagnosis method of a data driven multivariable control system. The method comprises the following steps: according to closed loop operation data, estimating white noise, identifying an interference model, and acquiring global process model residual errors and process model residual errors of different output channels; acquiring global model quality indexes according to the white noise and the global process model residual errors, detecting overall process model mismatching of a multivariable control system according to the global model quality indexes, acquiring overall model quality indexes and partial model quality indexes according to the white noise and the process model residual errors of different output channels, confirming input/output sub-channels of model mismatching of the multivariable control system, and judging that a process model of an input/output sub-channel has serious model mismatching if a partial model quality index isapproximate to 0. By adopting the method, process model mismatching can be diagnosed under a normal operation condition in the industrial process, system maintenance cost can be lowered, and system security can be improved.