Voluntary assisted partner notification (VAPN) services that use contract, provider, or dual referral modalities may be efficient to identify individuals with undiagnosed HIV infection. We aimed to assess the relative effectiveness of VAPN modalities in identifying undiagnosed HIV infections. VAPN was piloted in 23 health facilities in Kigali, Rwanda. We identified individuals with a new HIV diagnosis before antiretroviral therapy initiation or individuals on antiretroviral therapy (index cases), who reported having had sexual partners with unknown HIV status, to assess the association between referral modalities and the odds of identifying HIV-positive partners using a Bayesian hierarchical logistic regression model. We adjusted our model for important factors identified through a Bayesian variable selection. Between October 2018 and December 2019, 6336 index cases were recruited, leading to the testing of 7690 partners. HIV positivity rate was 7.1% (546/7690). We found no association between the different referral modalities and the odds of identifying HIV-positive partners. Notified partners of male individuals (adjusted odds ratio 1.84; 95% credible interval: 1.50 to 2.28) and index cases with a new HIV diagnosis (adjusted odds ratio 1.82; 95% credible interval: 1.45 to 2.30) were more likely to be infected with HIV. All 3 VAPN modalities were comparable in identifying partners with HIV. Male individuals and newly diagnosed index cases were more likely to have partners with HIV. HIV-positive yield from index testing was higher than the national average and should be scaled up to reach the first UNAIDS-95 target by 2030.