The purpose of this study was to determine how well self-report (questionnaire=QR) and trained observer (checklist=OBS) data recording methods compared with more expensive video analysis (VID) for estimating various peak physical loading exposure variables on the low backs of 99 employees during work in an automobile assembly plant. The variables studied were L4/L5 spine compression and shear forces, L4/L5 moment, trunk angle, and hand load. Peak low back loads associated with the working postures of, and the applied loads on, each worker were estimated using a 2D biomechanical model that could accommodate inertial forces acting in various directions on the hands independently. Correlations between the VID and OBS methods were greater for each variable than between VID and QR methods, with ranges in coefficients from 0.6 to 0.8, and 0.1 to 0.4, respectively, giving a discouraging impression of the QR, and the OBS method to a lesser degree, for peak low back exposure assessment. Despite the better performance of OBS method for individuals, it was still only able to account for between 36% and 64% of the variance relative to the VID method. When all workers were considered as a single group, compression and shear forces, moment and hand load estimates were the same regardless of method used to collect the data. Self-reported trunk flexion was significantly greater than that reported by trained observers or on video ( p<0.0001). Relevance to industry Considerable time and expense could be saved in large scale studies if it were possible to rely on worker's reports or observation of the physical demands of their jobs instead of traditional video and biomechanical analyses. Assessments of peak exposure of individuals using the self-report and observation methods were discouraging. Analysis of a single group proved more promising, but other groups need to be studied. Interview assisted self-reports may help to improve assessments of individuals and also need to be investigated in the future.