Impedance plethysmography (IPG) and compression ultrasonography (CUS) have been reported to be highly accurate for the diagnosis of deep vein thrombosis (DVT) in symptomatic patients. In many centres CUS has become the method of choice. However, direct comparisons of the accuracy of IPG to CUS have not been performed. To determine the test of choice we performed a two centre prospective comparison of IPG and CUS, with venography, and determined how the size and distribution of thrombi influenced the accuracy of each test. 495 symptomatic outpatients with suspected DVT had evaluable venograms. The prevalence of DVT was 27% (130/495), 84% (109) of which were proximal. The sensitivity of IPG and CUS for proximal vein thrombosis was 77% and 90% respectively (p = .002). The specificity of IPG was 93% whereas the specificity of CUS was 98% (p = 0.04). There were significant differences in accuracy between the two centres as a consequence of differences in the size and location of thrombi. The majority of proximal thrombi not detected by IPG and CUS involved less than 5 cm of the distal half of the popliteal vein and most of these thrombi occurred in one centre. Exclusion of these thrombi from the analysis increases the sensitivity of CUS to 99% (86/87) and IPG to 91% (72/79), for proximal thrombi (p = .019). The positive predictive value of CUS was strongly influenced by the number of abnormal venous segments (three sites were examined); 100% (80/80) if two or three sites were abnormal, but only 68% if a single site was involved. We conclude that: 1) CUS is more accurate than the IPG for the diagnosis of DVT in symptomatic outpatients, and this relationship holds true regardless of the size or location of the DVT, 2) the sensitivities of IPG and CUS are much lower for small proximal DVT, and 3) confirmatory venography is warranted if the abnormality with CUS is limited to one venous segment.