We analysed follow-up data from 18 486 primary total hip arthroplasties performed between 1967 and 2001 to assess the validity of clinical procedures in diagnosing loosening of prosthetic components. Sensitivity, specificity and predictive values were estimated with the radiological definition of loose or not loose as the ‘gold standard’. The prevalence of acetabular loosening increased from 0.6% to 13.9% during the period of the study and that of femoral loosening from 0.9% to 12.1%. Sensitivities and positive predictive values were low, suggesting that clinical procedures could not replace radiological assessment in the identification of loose prostheses. Specificities and negative predictive values were constantly above 0.86. The possibility of there being a prosthesis which is not loose in asymptomatic patients was consequently very high, particularly during the first five to six years after operation. The necessity of periodic clinical and radiological follow-up examinations of asymptomatic patients during the first five to six years after operation remains questionable. Symptomatic patients, however, require radiological assessment.