Under the analysis of classical decoupling methods, a new kind of decoupling method is proposed based on kernel methods. We consider the application of support vector regression and kernel ridge regression in multivariable decoupling design. Simulation results are presented to show the multivariable control system adopting the kernel method compensator can decouple the coupling effects among those parameters. A comparison of support vector regression and kernel ridge regression in decoupling performance is also discussed. The method is relatively simple and is easy to implement All these characteristics make the decoupled control system easy to use in a practical environment.