To verify that maximization of endurance is important among the functional criteria determining trunk muscle activation patterns, symmetric and asymmetric exertions were simulated using a detailed model consisting of 114 muscle slips crossing the lumbosacral junction and employing a cost function which maximizes endurance. First, the question whether meaningful comparisons can be made between activity predictions for individual muscle slips and surface EMG data recorded from larger anatomical entities was addressed. This was answered affirmatively, since activation patterns predicted by a coarse and a middle version of the model, in which activation was constrained to be equal within 14 or 32 groups of muscle slips, were similar to those predicted with each muscle slip controlled independently. Median correlation coefficients between activity vectors predicted by the simplified models and the detailed model were 0.88 and 0.97, respectively. The coarse model underestimated the endurance capacity by a median of 21%, the middle model by only 0.7%. Second, predicted activities within anatomical entities defined at this level of detail were compared to reference data derived from the literature (Lavender et al. 1992, Human Factors 34, 239-247; 1992, Journal of Orthopaedic Research 10, 691-700; Vink et al., 1988, Electromyography and Clinical Neurophysiology 28, 517-525). The predicted activity patterns of the erector spinae, external oblique and rectus abdominis muscles closely resembled the EMG patterns (r2 = 0.48-0.99). Furthermore, the observed distribution of activity between parts of the erector spinae muscle was adequately predicted.