The smaller index to ring finger (2D:4D) ratio has been considered as a 'male finger pattern' and is associated with sporting ability and a number of conditions. However, the ratio may vary according to what is measured, the hand selected and the method used. This study aimed to determine: (1) which bones (phalanges, metacarpals or both) account for variation in the 2D:4D ratio; (2) whether the ratio shows right-left symmetry or relates to hand dominance; and (3) the correlation between visual classification and measured determinations of the ratio based on radiographs. Hand radiographs obtained as part of a large osteoarthritis genetic study were examined. Each hand was classified visually into three types according to the relative length of the index and ring finger: Type 1 (index longer than ring), Type 2 (index = ring) and Type 3 (index shorter than ring). For both index and ring fingers we measured (1) from base of proximal to tip of distal phalanx and (2) metacarpal length. Reproducibility of the classification and measurements were examined using kappa and intraclass correlation coefficient; symmetry between left and right hands was examined using Bland and Altman's agreement analysis; and correlation between visual classification and 2D:4D ratio data was analysed using the anova linearity test. Data were obtained from 3172 radiographs (1636 men, 1536 women; mean age 67 +/- 7.9 years, range 45-86 years). Prevalence of Type 3 hand was 61% in men and 37% in women (P < 0.001). Men had smaller 2D:4D ratios than women for phalanges (0.908 versus 0.922, P < 0.01), metacarpals (1.152 versus 1.157, P < 0.01) and the sum of phalanges plus metacarpals (1.005 versus 1.015, P < 0.01). The mean difference between right and left was -0.001 (95% limit of agreement -0.035, 0.032) for the phalangeal ratio and 0.003 (95% limit of agreement -0.051 to 0.057) for the metacarpal ratio. The 2D:4D ratio did not associate with handedness or age. There was a linear trend between the visual classification of hand type and the 2D:4D ratio data (P < 0.001). More technical difficulties (due to positioning, finger trauma, osteoarthritis) were encountered with the phalangeal ratio and visual categorization than with the metacarpal ratio: the latter could be measured in 98.7% of the study population. We concluded that measured 2D:4D ratios and visual categorization can be derived from hand radiographs. The phalanges and metacarpals both contribute to the variation in 2D:4D ratio with smaller ratios observed in men than in women. The ratio is symmetrical with only very small differences between right and left hands. Visual classification may be a useful simple tool for future epidemiological studies but is more prone to bias from positioning than direct measurement. If radiographs are used for this purpose, we recommend the metacarpal ratio with measurement of a single index hand or an average of both as it is least affected by bias from malpositioning, trauma or common joint disease.