Respondent Driven Sampling study (RDS) is a population sampling method developed to study hard-toreach populations. A sample is obtained by chain-referral recruitment in a network of contacts within the population of interest. Such self-selected samples are not representative of the target population and require weighing observations to reduce estimation bias. Recently, the Network Model-Assisted (NMA) method was described to compute the required weights. The NMA method relies on modeling the underlying contact network in the population where the RDS was conducted, in agreement with directly observable characteristics of the sample such as the number of contacts, but also with more difficult-to-measure characteristics such as homophily or differential characteristics according to the response variable. Here we investigated the use of the NMA method to estimate HIV prevalence from RDS data when information on homophily is limited. We show that an iterative procedure based on the NMA approach allows unbiased estimations even in the case of strong population homophily and differential activity and limits bias in case of preferential recruitment. We applied the methods to determine HIV prevalence in men having sex with men in Brazilian cities and confirmed a high prevalence of HIV in these populations from 3.8% to 22.1%. Respondent-driven sampling (RDS) is a method to sample hard-to-reach populations such as injecting drug users, men who have sex with men (MSM), and sex workers 1. It uses chain-referral sampling, building on the underlying contact network for recruitment of participants. RDS starts by selecting seed individuals from the population of interest. They receive a fixed number of coupons to distribute to individuals in their contact network who meet certain eligibility criteria. In turn, individuals receiving a coupon recruit new participants among their contacts, leading to successive recruitment waves until the target number of individuals for the survey is reached 2. A drawback of the method is that the final sample is not representative of the target population, introducing bias in naïve estimates of, say, prevalence. Statistical procedures using generalized Horvitz-Thompson estimators can reduce biases 3,4. In those, weights are computed according to referral patterns, estimated network size, number of ties between subgroups of interest 5 , differences in the number of partners declared by participants 6 or homophily in chain referrals 7. Weights can also be computed using bootstrap procedures 8. Yet, numerous issues affect the reliability and validity of RDS estimates because several hypotheses are required 3,9-11 : (1) the population size need to be large compared to the RDS sample; (2) sampling must occur with replacement; (3) population homophily should be weak; (4) seeds should be selected at random. Recently, the network model-assisted (NMA) method has been shown to increase the robustness in prevalence estimation from RDS data with respect to these assumptions 11. In this approach, characterist...