This study examines the reliability of quantitative and qualitative muscle composition measurements of paraspinal muscle cross-sectional areas (CSAs) from routine lumbar spine magnetic resonance images and their association with maximal isokinetic lifting performance. The extent of paraspinal muscle composition reflects back function is currently not known. Measurements were repeated 4-8 weeks apart and different measurements of related constructs were compared. Participants were a population-based sample of 169 males, 35-67 years old, without considering the presence or absence of a history of low back pain or related problems in the selection of subjects. The quantitative and qualitative muscle composition measurements for axial magnetic resonance (MR) images of paraspinal muscles at the L3-L4 lumbar spine level, isokinetic lifting force and work, and body fat percentage were the main outcome measures. Results showed that the reproducibility of different paraspinal muscle composition measurements at the L3-L4 level was excellent for CSAs (ICC=0.95-0.99) and quantitative muscle composition measurements using cerebrospinal fluid adjusted signal intensity (ICC=0.96-0.99), and moderate for qualitative muscle composition ratings (Kappa=0.54-0.76). The correlations of the quantitative and qualitative muscle composition measurements with isokinetic lifting force and work were generally low (r=0.02-0.41), and favoured the qualitative assessments. In conclusion, quantitative and qualitative muscle composition measurements of paraspinal muscles are highly reproducible tissue measures, have low associations with body fat and isokinetic lifting performance, and show that paraspinal muscle morphology using routine spine magnetic resonance imaging (MRI) is poorly related to back function.