Randomization is recognized as a tool for removing bias in any experiment. In particular, in Epidemiologic treatment group comparisons, this tool is considered as a gold standard. However, in a practical scenario, at times, the experiments may be poorly designed, or may not be executed properly, and therefore, bias may creep in. Also, having access to huge amount of non-randomized observational hostital data forces us to develop statistical framework of utilizing non-randomized data. Various approaches had been suggested in the literature to ensure comparability of such treated and untreated groups. The approach of propensity scores is one way to adjust for these bias sources in absense of randomization. In this talk, I will discuss briefly the main concepts of how propensity scores are utilized, implemented (in theory and in R/Stata/SAS) and interpreted in these type of situations, with special emphasis on Epidemiologic data.