Objective. To assess inter-observer agreement in the interpretation of renal angiograms. Design. Comparison of the assessment of renal angiograms by three experienced radiologists, who evaluated the number of renal arteries and the presence, location, aspect and severity of a renal artery stenosis. Setting. General hospital and university hospital serving urban and rural populations. Patients. Patients with difficult-to-treat hypertension referred for diagnostic work-up; 312 angiograms with the intra-arterial digital subtraction technique were obtained from 289 consecutive patients. Main outcome measures. Inter-observer agreement was tested for the following parameters: number of arteries per kidney, presence of stenosis, location of stenosis (truncal, ostial), aspect of stenosis (concentric, eccentric, post-stenotic dilatation), severity of stenosis (reduction of lumen diameter in categories of 30%, 40%, etc. to 100%), and overall quality of the angiographic images. Kappa (κ) values and weighted κ between the three pairs of radiologists were used as estimates of inter-observer agreement. Results. Agreement about the number of renal arteries was reasonable (κ = 0.50-0.72), as was agreement about the presence of stenosis (κ = 0.68-0.86). Agreement about stenosis location and aspect was poor (κ = 0.26-0.47 and κ = 0.15-0.26, respectively). There was general agreement about the severity of stenosis (weighted κ = 0.65-0.70), but it was not possible to distinguish between 50 and 60% stenosis or between 60 and 70% stenosis (κ < 0.40). No correlation was found between agreement on severity of stenosis and the quality of the images. Conclusions. It is not realistic to make statements about what degree of renal artery stenosis is clinically significant, as long as the intra-arterial angiogram with digital subtraction remains the gold standard. It is likewise risky to rely too strongly on stenosis morphology as visualized by renal angiography in choosing between balloon angioplasty and stent deployment.