Seventeen drawings of localised low-back pain were analysed by two assessors using 4 systems. Three were grid-based systems and one was by computer. The mean area or 'extent' was calculated to be 7.7%, 4.7%, 3.6% and 2.3% of the body outline using 45, 200, 560 and 61,102 section analyses, respectively. The computer-assisted method provided a significantly lower estimate of pain extent than the grid-based assessments as expected. Analysis of variance showed that the method of analysis provided greater source of variation than raters (P < 0.0001). Inter-rater reliability was high using all 4 systems calculated using intraclass correlation and the kappa statistic. Correlation coefficients of extent between the systems varied from 0.46 to 0.94. Correlation was highest between systems of adjacent magnitude of sections. It is concluded that grid-based assessment of small areas overestimates the actual area of pain and this may account for the lack of sensitivity to change in clinical status.