This cross-sectional, prospective study aimed to produce normal reference data for measurements of the lumbar multifidus muscle. A total of 120 subjects, 68 females (aged 20-64 years) and 52 males (20-69 years) were studied. Bilateral transverse ultrasound images were made of multifidus at the fourth and fifth lumbar vertebrae (L4 & L5). Cross-sectional area (CSA, cm(2)) and linear dimensions (AP, anteroposterior; Lat, lateral) were measured and the latter expressed as a ratio (AP/Lat) to reflect shape. Relationships between CSA and anthropometric measures were examined. Multifidus CSA was larger in males (P<0.001) and age had no effect. The CSA was larger at L5 than L4 (P<0.001) and highly correlated between the two levels (males r=0.82, females 0.80). Differences in muscle shape were observed for gender, age and vertebral level. Between-side symmetry was high for size but not shape (CSA <10% difference). Linear measurements multiplied (APxLat) correlated highly with CSA (all groups r0.94, P<0.0001). The AP dimension was also acceptably predictive of CSA at L4 (r0.79). There were no clinically useful correlations between CSA and anthropometric measures. These findings provide normal references ranges for objective assessment of lumbar multifidus. This paper also addresses specific practical issues when scanning multifidus.