The detection of galactomannan antigen in urine was investigated in 26 bone marrow transplant recipients using an Aspergillus latex agglutination test (Pastorex). After modification of the method, which was originally devised for serum testing, the detection limit in native urine was approximately 20 ng/ml. Antigen was found in 79 (36.4%) of 217 serial urine samples, compared to 40 (11.8%) of 340 serum samples. As a rule, antigenuria preceded antigenemia and was more persistent. The sensitivity, specificity, positive predictive value and negative predictive value of antigenuria for autopsy-proven aspergillosis and clinically suspected Aspergillus infection were 57%, 53%, 31% and 77%, respectively, while those of antigenemia were 43%, 53%, 25% and 71%. It is concluded that urine testing is more reliable than serum testing for the detection of Aspergillus galactomannan. The detection of antigen, however, whether in serum or in urine, allows no clear distinction between Aspergillus infection and exposure to non-infectious Aspergillus antigens.