This correspondence considers a MIMO relay downlink system using precoding and limited feedback. Since the conventional precoding matrices are directly obtained by treating the quantized channel state information (CSI) feedback as real CSI, we propose an optimized relay precoding strategy by taking both effects of channel quantization error and feedback delay into account. Conditioned on the outdated CSI quantization available at the relay station, the relay precoding is optimized via minimizing an expected mean-square error (MSE) criterion over CSI mismatches. By using Karush-Kuhn-Tucker (KKT) conditions, a closed-form solution to the relay design is achieved with comparable computational complexity relative to the conventional approach. Numerical results verify the effectiveness of the proposed scheme.