Few studies have been carried out on the changes in biomechanical loading on low-back tissues during prolonged lifting. The purpose of this paper was to develop a model for continuously estimating erector spinae muscle loads during repetitive lifting and lowering tasks. The model was based on spine kinematics and bilateral lumbar and thoracic erector spinae electromyogram (EMG) signals and was developed with the data from eight male subjects. Each subject performed a series of isometric contractions to develop extensor moments about the low back. Maximum voluntary contractions (MVCs) were used to normalize all recorded EMG and moment time-histories. Ramp contractions were used to determine the non-linear relationship between extensor moments and EMG amplitudes. In addition, the most appropriate low-pass filter cut-off frequencies were calculated for matching the rectified EMG signals with the moment patterns. The mean low-pass cut-off frequency was 2.7 (0.4) Hz. The accuracy of the non-linear EMG-based estimates of isometric extensor moment were tested with data from a series of six rapid contractions by each subject. The mean error over the duration of these contractions was 9.2 (2.6)% MVC. During prolonged lifting sessions of 20 min and of 2 h, a model was used to calculate changes in muscle length based on monitored spine kinematics. EMG signals were first processed according to the parameters determined from the isometric contractions and then further processed to account for the effects of instantaneous muscle length and velocity. Simple EMG estimates were found to underestimate peak loading by 9.1 (4.0) and 25.7 (11.6)% MVC for eccentric and concentric phases of lifting respectively, when compared to load estimates based on the mechanically corrected EMG. To date, the model has been used to analyze over 5300 lifts.