OBJECTIVES--To determine the extent to which variation in rates of referral among general practitioners may be explained by inappropriate referrals and to estimate the effect of implementing referral guidelines. SETTING--Practices within Cambridge Health Authority and Addenbrooke's Hospital, Cambridge. MAIN OUTCOME MEASURES--Data on practice referral rates from hospital computers, inappropriate referrals as judged by hospital consultants, and inappropriate referrals as judged against referral guidelines which had been developed locally between general practitioners and specialists. Effect of referral guidelines on referral patterns as judged by general practitioners using the guidelines in clinical practice. RESULTS--There was 2.5-fold variation in referral rates among general practices. According to the specialists, 9.6% (95% confidence interval 6.4% to 12.9%) of referrals by general practitioners and 8.9% (2.6% to 15.2%) of referrals from other specialists were judged possibly or definitely inappropriate. Against locally determined referral guidelines 15.9% of referrals by general practitioners were judged possibly inappropriate (11.8% to 20.0%). Elimination of all possibly inappropriate referrals could reduce variation in practice referral rates only from 2.5-fold to 2.1-fold. An estimate of the effect of using referral guidelines for 60 common conditions in routine general practice suggested that application of guidelines would have been unlikely to reduce rates of referral in hospital (95% confidence interval -4.5% to 8.6% of consultations resulting in referral). CONCLUSION--The variation in referral rates among general practitioners in Cambridge could not be explained by inappropriate referrals. Application of referral guidelines would be unlikely to reduce the number of patients referred to hospital.