This article considers regression problems where both the predictor and the response are functional in nature. Driven by the desire to build a parsimonious model, we consider functional reduced rank regression in the framework of reproducing kernel Hilbert spaces, which can be formulated in the form of linear factor regression with estimated multivariate factors, and achieves dimension reduction in both the predictor and the response spaces. The convergence rate of the estimator is derived. Simulations and real datasets are used to demonstrate the competitive performance of the proposed method.