In a prospective study of 1,805 hospitalized patients in the Stroke Data Bank of the National Institute of Neurological and Communicative Disorders and Stroke, the 1,273 with infarction were classified into diagnostic subtypes. Diagnosis was based on the clinical history, examination, and laboratory tests including computed tomography, noninvasive vascular imaging, and where safe and relevant, angiography. Five hundred and eight cases (fully 40%) were labeled as infarcts of undetermined cause (IUC), of which 138 (27%) were evaluated with both computed tomography and angiography. The clinical syndrome and computed tomographic and angiographic findings in 91 (65.9%) of these 138 IUC cases were clearly not attributable to large-artery thrombosis and could permit reclassification of the infarct as due to some form of embolism. Failure to define a source of embolus kept them in the category of IUC. Thirty-one cases (22.5%) could be reclassified as due to stenosis or thrombosis of a large artery, and 16 (11.6%) as lacunar infarction. To determine if those selected for angiography among the IUC patients differed from those with other final diagnoses, a stepwise multiple logistic model was used. The most important characteristics were young age, presence of a superficial infarct, prior transient ischemic attack, low weakness score, and presentation with a nonlacunar syndrome. The results of the model suggest that angiography use was determined by clinical characteristics uniformly across centers and not by final diagnosis. Continued use of the category IUC may help clarify risk factors and stroke subtypes, allow new mechanisms of ischemic stroke to be uncovered, and prevent classification categories of stroke used in clinical trials from becoming too broad.