Our paper compares the L-moments (MLM), TL-moments (MTLM), and maximum likelihood estimation (MLE) methods in order to select the best-fit distribution of annual maximum stream flow (AMSF) data for at-site flood frequency analysis (FFA) in Pakistan. Initially this study considered different probability distributions. Best distribution for each site is identified using different goodness of fit tests such as mean absolute deviation index (MADI), Anderson darling (AD) test, probability plot correlation coefficient (PPCC), and L-moments ratio diagram. Results show that GPA distribution is the most suitable distribution for most of the sites, followed by GLO and GEV distributions, respectively. MLM is found to be the most suitable estimation method in finding the best-fit distribution for most of the sites in this study, followed by MTLM and MLE. For at-site FFA we also estimated different return periods associated with given flood magnitudes (quantiles of best fit distribution/maximum annual discharge values). It is found that estimated flows based on fitted distribution are in close agreement with observed flows.