The objective of this study was to use and evaluate three postural assessment methods for epidemiological studies of back disorders. The methods were: (1) a data-logging inclinometer; (2) observations by trained observers; (3) self-reports by employees. All methods were feasible in 50 heavy industry worksites. Inclinometry provided quantitative measures of flexion-extension (mean 17 degrees, SD 11.2 degrees), lateral flexion (mean 8.5 degrees, SD 2.6 degrees) and trunk movement speed (mean 14.3 degrees per second, SD 4.9 degrees per second). Observations and self-reports provided estimates of time spent in various trunk angles, general postures, materials handling and vehicles. Compared to observations, self-reports under-reported less common tasks, but over-reported task durations. In statistical modelling to determine if observations or self-reports could be used to estimate measured postures, observations accounted for 30 to 61% of the inclinometer measurement variance and self-reports for 33 to 40%. A combination of inclinometry and observations would be an ideal option to provide both depth and breadth of data on postures and other physical exposures for epidemiological research.