BACKGROUND: Giant cell arteritis (GCA) is a relatively common form of primary systemic vasculitis, which, if left untreated, can lead to permanent sight loss. We compared ultrasound as an alternative diagnostic test with temporal artery biopsy, which may be negative in 9-61% of true cases. OBJECTIVE: To compare the clinical effectiveness and cost-effectiveness of ultrasound with biopsy in diagnosing patients with suspected GCA. DESIGN: Prospective multicentre cohort study. SETTING: Secondary care. PARTICIPANTS: A total of 381 patients referred with newly suspected GCA. MAIN OUTCOME MEASURES: Sensitivity, specificity and cost-effectiveness of ultrasound compared with biopsy or ultrasound combined with biopsy for diagnosing GCA and interobserver reliability in interpreting scan or biopsy findings. RESULTS: We developed and implemented an ultrasound training programme for diagnosing suspected GCA. We recruited 430 patients with suspected GCA. We analysed 381 patients who underwent both ultrasound and biopsy within 10 days of starting treatment for suspected GCA and who attended a follow-up assessment (median age 71.1 years; 72% female). The sensitivity of biopsy was 39% [95% confidence interval (CI) 33% to 46%], which was significantly lower than previously reported and inferior to ultrasound (54%, 95% CI 48% to 60%); the specificity of biopsy (100%, 95% CI 97% to 100%) was superior to ultrasound (81%, 95% CI 73% to 88%). If we scanned all suspected patients and performed biopsies only on negative cases, sensitivity increased to 65% and specificity was maintained at 81%, reducing the need for biopsies by 43%. Strategies combining clinical judgement (clinician's assessment at 2 weeks) with the tests showed sensitivity and specificity of 91% and 81%, respectively, for biopsy and 93% and 77%, respectively, for ultrasound; cost-effectiveness (incremental net monetary benefit) was £485 per patient in favour of ultrasound with both cost savings and a small health gain. Inter-rater analysis revealed moderate agreement among sonographers (intraclass correlation coefficient 0.61, 95% CI 0.48 to 0.75), similar to pathologists (0.62, 95% CI 0.49 to 0.76). LIMITATIONS: There is no independent gold standard diagnosis for GCA. The reference diagnosis used to determine accuracy was based on classification criteria for GCA that include clinical features at presentation and biopsy results. CONCLUSION: We have demonstrated the feasibility of providing training in ultrasound for the diagnosis of GCA. Our results indicate better sensitivity but poorer specificity of ultrasound compared with biopsy and suggest some scope for reducing the role of biopsy. The moderate interobserver agreement for both ultrasound and biopsy indicates scope for improving assessment and reporting of test results and challenges the assumption that a positive biopsy always represents GCA. FUTURE WORK: Further research should address the issue of an independent reference diagnosis, standards for interpreting and reporting test results and the evaluation of ultrasound training, and should also explore the acceptability of these new diagnostic strategies in GCA. FUNDING: The National Institute for Health Research Health Technology Assessment programme.