OBJECTIVE: To study hospital admission rates for fractures of the proximal femur over a period when incidence is reported to have increased, compensating for known lack of precision in coding, excluding nonemergency admissions and transfers, and modelling for age, period, and cohort effects. DESIGN: Validation of coding of a sample of hospital admissions followed by study of two sets of routinely collected statistical abstracts of hospital records; graphical analysis and statistical modelling were used to search for period and cohort effects. SETTING: Oxfordshire and west Berkshire in 1968-86, covered by the Oxford record linkage study (ORLS), and ENGLAND in 1968-85, covered by the hospital inpatient enquiry (HIPE). The ORLS and HIPE datasets are almost independent (ORLS contributed about 1.8% of the HIPE data). SUBJECTS: Records of patients aged 65 and over. OUTCOME MEASURES: Admission rates for fractured neck of femur and fracture of other and unspecified parts of femur (N820 and N821), and evidence of period and cohort effects. RESULTS: The validation study indicated that it was important to combine the codes 820 and 821 in this age group. Admission rates increased over the period studied in both HIPE and ORLS datasets. In HIPE the pattern was of two plateaux separated by a period of rapid rise in the late 1970s. In the ORLS data there was a more steady rise. Statistical analysis showed significant period and cohort effects but much of this was attributable to the component of the model common to both period and cohort effects (termed "drift"). CONCLUSIONS: The finding that admission rates increased in both datasets, combining relevant codings and restricting analysis to emergency admissions, strongly suggests that the rise was real. At least part of the period effect in the HIPE data, however, might be attributable to a sampling artefact. The cohort effect in incidence rates of femoral fracture has not been previously shown and would be compatible with a number of aetiological hypotheses.