This paper presents a TSK-type recurrent fuzzy neural network (TRFNN) system and hybrid algorithm to control nonlinear uncertain systems. The TRFNN is modified from the RFNN to obtain generalization and fast convergence rate. The consequent part is replaced by linear combination of input variables and the internal variable- fire strength is feedforward to output to increase the network ability. Besides, a hybrid learning algorithm (GA_BPPSO) is proposed to increase the convergence, which combines the genetic algorithm (GA), back-propagation (BP), and particle swarm optimization (PSO). Several simulation results are proposed to show the effectiveness of TRFNN system and GA_BPPSO algorithm.