Central venous access is extensively used in oncology, though practical information from randomized trials on the most convenient insertion modality and site is unavailable. Four hundred and three patients eligible for receiving i.v. chemotherapy for solid tumors were randomly assigned to implantation of a single type of port (Bard Port™, Bard Inc., Salt Lake City, UT), through a percutaneous landmark access to the internal jugular, a ultrasound (US)-guided access to the subclavian or a surgical cut-down access through the cephalic vein at the deltoid–pectoralis groove. Early and late complications were prospectively recorded until removal of the device, patient's death or ending of the study. Four hundred and one patients (99.9%) were assessable: 132 with the internal jugular, 136 with the subclavian and 133 with the cephalic vein access. The median follow-up was 356.5 days (range 0–1087). No differences were found for early complication rate in the three groups {internal jugular: 0% 95% confidence interval (CI) 0.0% to 2.7%, subclavian: 0% (95% CI 0.0% to 2.7%), cephalic: 1.5% (95% CI 0.1% to 5.3%)}. US-guided subclavian insertion site had significantly lower failures (e.g. failed attempts to place the catheter in agreement with the original arm of randomization, P=0.001). Infections occurred in one, three and one patients (internal jugular, subclavian and cephalic access, respectively, P=0.464), whereas venous thrombosis was observed in 15, 8 and 11 patients (P=0.272). Central venous insertion modality and sites had no impact on either early or late complication rates, but US-guided subclavian insertion showed the lowest proportion of failures.