Clinical significance of lumbar lordosis has not been agreed on. Our purpose is to compare lordotic measurements of normal and pain subjects and to test the validity of a new anthropometric model of lumbar curvatures. Digitized radiographic points (body corners) from standing lateral lumbar radiographs were modeled with ellipses in a least-squares method and were used to create segmental angles, a global angle at L1-L5, a Cobb angle from T12 to S1, Ferguson's sacral base angle, and an angle of pelvic tilt. Fifty normal subjects were matched in age, sex, weight, and height with 50 acute pain subjects, 50 chronic pain subjects, and 24 pain subjects with radiographic abnormalities. Of 11 angles, 2 distances, and 2 ratios, statistical analysis was significantly different across groups for 12 of these measurements, with the alternative hypotheses accepted for the other 3 measurements. The lordosis of both normal and low back pain subjects can be successfully modeled with a portion (approximately 86 degrees) of an ellipse, but with different major and minor axis ratios. The normal group's average elliptic lordosis has the smallest least-squares error, approximately 1 mm per digitized point, with (minor axis)/(major axis) ratio = 0.39, L1-L5 global angle = 40 degrees, and Cobb angle = 65 degrees. The chronic and radiographic abnormalities pain groups have an elongated ellipse with hypolordosis, reduced L1-L5 global angle = 29.6-35 degrees, reduced Cobb angle = 57-58 degrees, and elliptic axis ratio = 0.27-0.30. The acute pain group is hyperlordotic with the largest L1-L5 global angle, largest Cobb angle = 70 degrees, largest Ferguson's angle, and largest pelvic tilt angle.