The need to obtain unbiased information among hard-to-reach and hidden populations for behavioural and biological surveillance, epidemiological studies, and intervention program evaluations has led researchers to search for a suitable sampling method. One method that has been tested among IDU and MSM recently is respondent-driven sampling (RDS). We used RDS to conduct a behavioural survey among FSWs and MSM in two urban centres in Papua New Guinea (PNG). In this paper we present the lessons learned implementing RDS in a developing country setting. We also present comparisons of RDSAT-adjusted versus unadjusted crude estimates of some key socio-demographic indicators as well as comparisons between the estimates from RDS and a hypothetical time-location sample (TLS). Overall, the use of RDS among the MSM and FSWs in PNG had numerous advantages in terms of collecting a required sample size in a short time period, minimizing costs and maximising security for staff and respondents. Although there were a few problems these were easily remedied and we would recommend RDS for other similar studies in PNG and other developing countries.