An online tuning method based on a partial least squares (PLS) decomposition structure for multiloop proportional-integral-derivative controllers in nonlinear processes is proposed. Unlike traditional decoupling design, the proposed dynamic PLS (DynPLS) model derived from PLS and the linear dynamic model can decompose a multi-input multi-output (MIMO) process into a multiloop control system in a reduced subspace. To update DynPLS in the nonlinear process, the instantaneous linearized neural network model at each sampling time is used to extract the linear dynamic part of DynPLS for quickly updating the current model. Because of the decomposition of a complex MIMO process, this scheme takes advantage of the simplicity to enhance feasibility. Simulation case studies are used to demonstrate the effectiveness of the control design procedures of nonlinear MIMO processes.