Respondent-driven sampling (RDS) is nowadays a well-known sampling method. However, although it is able to find members of hard-to-reach populations, the estimates generated by it remain a point of concern. The evaluation of those estimates is difficult due to the lack of knowledge about the actual parameters from the vast majority of hidden populations, worldwide. One additional problem of RDS-based studies is associated with the dependency on the information about the contact network (degree) of each individual. The aim of this work is to test RDS estimators in populations simulated with diverse characteristics. Four populations were simulated using different random graph models (Erdös-Renyi – ER; Watts-Strogats – WS; Barabasi-Albert – BA; Interconnected-Islands – II). Each population was "infected" using four different transmission processes. From each combination of population x transmission, one thousand samples with 500 individuals each were obtained using a RDS-like method and three estimators were used to predict population prevalence of the "infection": RDS-I, RDS-II, and the classical frequentist estimator. Several biases were purposively simulated, taking in consideration the information on “degree” used by RDS estimators. Also, the same number of samples were obtained using the standard random sampling method and the respective prevalence estimates were obtained using the classical frequentist estimator. Results were depicted as graphs aiming to visualize the effect of different biases in relation to the actual infection rate in each population, as well as the variance of the estimates. The most important effect observed was associated to the underlying connectivity of each simulated population. The Interconnected Islands model produced the greater variability and no RDS-based strategy could properly address the emerging biases. The degree bias produced only small effect on the results, especially when RDS estimators I/II were used. The best results for RDS-based samples were observed when the “infection” was randomly attributed, without any relation with the underlying network structure.