HIV antigen detection kits are available from a number of commercial sources. Abbott, Coulter, and Du Pont antigen kits were used to test 661 sera collected sequentially from 65 members of the Toronto Sexual Contact Study (TSCS). The sera had been collected at 3-month intervals over 4 years from nine persistently HIV-seronegative men, 14 seroconverters, and 42 seroprevalent participants. Antigen was not detected in any seronegative men. Two of 14 seroconverters were antigen positive in the specimen immediately preceding seroconversion (by all kits). Antigen was detected in 22 of 56 seropositive participants; of these, 16 of 22 demonstrated the emergence of antigen during observation. Discrepancies were noted in the time of detection of antigen (ranging from 3 months to more than 3 years) in nine participants. Although overall concordance among all kits for all specimens appears high (95.4%), when the bias introduced by testing multiple specimens from the same patient is removed, the lower bound of concordance among all three kits is estimated to be 80%. Similarly, after correction, the upper and lower bound of estimates of sensitivity are Abbott 96, 92%; Coulter 88, 63%; and Du Pont 88, 58%. There are significant differences in the performance characteristics of these commercial products for the detection of HIV antigen in serum.