INTRODUCTION: New HIV testing strategies are needed to reach the United Nations' 90-90-90 target. HIV self-testing (HIVST) can increase uptake, but users' perspectives on optimal models of distribution and post-test services are uncertain. We used discrete choice experiments (DCEs) to explore the impact of service characteristics on uptake along the testing cascade. METHODS: DCEs are a quantitative survey method that present respondents with repeated choices between packages of service characteristics, and estimate relative strengths of preferences for service characteristics. From June to October 2016, we embedded DCEs within a population-based survey following door-to-door HIVST distribution by community volunteers in two rural Zimbabwean districts: one DCE addressed HIVST distribution preferences; and the other preferences for linkage to confirmatory testing (LCT) following self-testing. Using preference coefficients/utilities, we identified key drivers of uptake for each service and simulated the effect of changes of outreach and static/public clinics' characteristics on LCT. RESULTS: Distribution and LCT DCEs surveyed 296/329 (90.0%) and 496/594 (83.5%) participants; 81.8% and 84.9% had ever-tested, respectively. The strongest distribution preferences were for: (1) free kits - a $1 increase in the kit price was associated with a disutility (U) of -2.017; (2) door-to-door kit delivery (U = +1.029) relative to collection from public/outreach clinic; (3) telephone helpline for pretest support relative to in-person or no support (U = +0.415); (4) distributors from own/local village (U = +0.145) versus those from external communities. Participants who had never HIV tested valued phone helplines more than those previously tested. The strongest LCT preferences were: (1) immediate antiretroviral therapy (ART) availability: U = +0.614 and U = +1.052 for public and outreach clinics, respectively; (2) free services: a $1 user fee increase decreased utility at public (U = -0.381) and outreach clinics (U = -0.761); (3) proximity of clinic (U = -0.38 per hour walking). Participants reported willingness to link to either location; but never-testers were more averse to LCT. Simulations showed the importance of availability of ART: ART unavailability at public clinics would reduce LCT by 24%. CONCLUSIONS: Free HIVST distribution by local volunteers and immediately available ART were the strongest relative preferences identified. Accommodating LCT preferences, notably ensuring efficient provision of ART, could facilitate "resistant testers" to test while maximizing uptake of post-test services.