This study investigates the multi-objective optimization of quality characteristics for a T300/epoxy prepreg tape-wound cylinder. The method integrates the Taguchi method, grey relational analysis (GRA) and response surface methodology, and is adopted to improve tensile strength and reduce residual stress. In the winding process, the main process parameters involving winding tension, pressure, temperature and speed are selected to evaluate the parametric influences on tensile strength and residual stress. Experiments are conducted using the Box–Behnken design. Based on principal component analysis, the grey relational grades are properly established to convert multi-responses into an individual objective problem. Then the response surface method is used to build a second-order model of grey relational grade and predict the optimum parameters. The predictive accuracy of the developed model is proved by two test experiments with a low prediction error of less than 7%. The following process parameters, namely winding tension 124.29 N, pressure 2000 N, temperature 40 °C and speed 10.65 rpm, have the highest grey relational grade and give better quality characteristics in terms of tensile strength and residual stress. The confirmation experiment shows that better results are obtained with GRA improved by the proposed method than with ordinary GRA. The proposed method is proved to be feasible and can be applied to optimize the multi-objective problem in the filament winding process.