Bridge health monitoring data often encounter missing values due to sensor failures and other factors. Existing data recov ery methods have not effectively utilized the temporal and spatial correlations in the data. In this paper， a multi-channel recovery method for bridge monitoring data based on temporal and spatial correlations is proposed. The original data is preprocessed using a Kalman filter to eliminate random errors. The preprocessed data is divided into training and testing sets， and training samples are constructed using a sliding window approach with masking. The data recovery issue is formulated as a time series prediction issue. Besides， an end-to-end LSTM network architecture is trained to leverage the temporal and spatial correlations in the historical data of the sensors which enables the recovery of missing data. The proposed method is validated using the measured deflection and ca ble force data from a suspension bridge， and the performance of single-channel and multi-channel data recovery is discussed. Com pared to the traditional RNN models， results show that the proposed method achieves a 22% improvement in accuracy when the data missing rate is 60%. Moreover， the method effectively utilizes the temporal and spatial correlations among different channels， enabling simultaneous recovery of data from multiple channels.