OBJECTIVE To assess the accuracy of serology to predict the presence of cytomegalovirus (CMV) in semen of homosexual men without and with HIV coinfection. METHODS Semen CMV was detected by electron microscopy and by polymerase chain reaction (PCR) amplification; paired serum was tested for CMV IgG/IgM. Semen HIV was detected by reverse transcription-PCR. METHODS Licensed clinical and research laboratory. METHODS Sixty-eight men. METHODS None. METHODS Frequency of CMV and HIV in semen. RESULTS Cytomegalovirus was detected by electron microscopy in 3 of 10 specimens examined. Forty-six (89%) of 52 HIV-infected men were seropositive for CMV by combined assay for IgG/IgM; two more (48 of 52, 92%) were seropositive for CMV IgG by separate assay; 25 (48%) of the HIV-infected men had PCR-detectable CMV DNA in at least one semen specimen, 22 of whom (42%) had CMV in all specimens. Nineteen (13%) of the 150 specimens tested positive for HIV, whereas 67 (45%) tested positive for CMV; seven specimens tested positive for both CMV and HIV. Cytomegalovirus, but not HIV, detection in semen correlated with decreased CD4(+) lymphocytes in peripheral blood (<700/μL) but was not accurately predicted by serology, leukocytospermia, or age. CONCLUSIONS Cytomegalovirus in semen is not accurately predicted by serology. Sperm banking needs to include direct assessment of CMV in semen specimens. Strategies to eliminate CMV from semen specimens are needed to alleviate the risk of virus transmission.