The invention discloses a method for directly predicting the wind-induced vibration response of a bridge based on long-term actually measured data. The method comprises the following steps: extracting primary information of long-term actually measured data of a structural health monitoring system (SHMS) of the bridge by utilizing a principal component analysis method at first, and constructing an SHMS actually measured data-bridge wind-induced vibration response model by adopting a BP neural network in combination with a parameter influencing the wind-induced vibration response of a bridge structure; then, inputting data, such as actually measured wind speeds and wind directions, into the model to obtain a wind-induced vibration response predication value, and comparing and checking the wind-induced vibration response predication value with an SHMS actually measured value of the bridge so as to obtain reliable wind-induced vibration response of the bridge; and finally, storing the latest actually measured data in an SHMS long-term actually measured database, and continuously optimizing a BP neural network prediction model so as to further improve the prediction precision of the wind-induced vibration response. With the help of the SHMS long-term actually measured data, the traditional finite element model is replaced by the neural network; and thus, an effective means and reliable analysis basis are provided for monitoring and safety assessment of wind-induced vibration of the bridge.