This paper proposes a Support Vector-trained Recurrent Fuzzy System (SV-RFS) which comprises recurrent Takagi- Sugeno (TS) fuzzy if-then rules. The SV-RFS memories past input information by feeding the past firing strength of a fuzzy rule back to itself. The rules are generated based on a clustering-like algorithm. The feedback loop gains and consequent part parameters are learned through support vector regression (SVR) in order to improve system generalization ability. The SV-RFS is applied to noisy chaotic sequence prediction to verify its effectiveness.