The Joint United Nations Programme on AIDS (UNAIDS) encourages governments to use the Modes of Transmission (MOT) model to estimate numbers of infections from various risks and thereby to guide HIV prevention efforts. A simple design error in the model – ignoring frequent sero-concordance among spouses – hugely inflates estimates of infections from spouses. Applications of the model to Uganda and Swaziland illustrate the impact of this error. For Uganda, using data on sero-discordance in couples cuts estimated annual spouse-to-spouse transmission from 60,948 to 32,100; with this change, the revised MOT model – considering infections from all risks – estimates only 52% of incidence needed to explain Uganda’s epidemic trajectory. For Swaziland, using data on sero-discordance in couples cuts annual spouse-to-spouse HIV transmission from 9,166 to 4,300, and the revised model estimates only 50% of infections needed to explain Swaziland’s epidemic trajectory. This error has similar impacts on MOT estimates for other African countries. Several design errors in the MOT model can be fixed. However, even with a revised design, the model’s dependence on unreliable data (numbers of risk events) and parameters (transmission per event) leads to unreliable estimates. To guide HIV prevention efforts, more reliable information about modes of transmission is and can be available from prospective studies of risks for incident infections and investigations that trace infections.