Low back loading during occupational lifting is thought to be an important causative factor in the development of low back pain. In order to regulate spinal loading in the workplace, it is necessary to measure it accurately. Various methods have been developed to do this, but each has its own limitations, and none can be considered a "gold standard". The purpose of the current study was to compare the results of three contrasting techniques in order to gain insight into possible sources of error to which each is susceptible. The three techniques were a linked segment model (LSM), an electromyographic (EMG)-based model, and a neural network (NN) that used both EMG and inertial sensing techniques. All three techniques were applied simultaneously to calculate spinal loading when eight volunteers performed a total of eight lifts in a laboratory setting. Averaged results showed that, in comparison with the LSM, the EMG technique calculated a 25.5+/-33.4% higher peak torque and the NN technique a 17.3+/-10.5% lower peak torque. Differences between the techniques varied with lifting speed and method of lifting, and could be attributed to differences in anthropometric assumptions, antagonistic muscle activity, damping of transient force peaks by body tissues, and, specific to the NN, underestimation of trunk flexion. The results of the current study urge to reconsider the validity of other models by independent comparisons.