The importance of the analysis and understanding of the network traffic has constantly been increasing due to insights that this provides towards determination of user behaviour and resource usage. The data analyses in order to determine the related parameters are performed by selection of a small subset of the complete flow data due to data privacy and heavy computational/memory load issues. That is, sampling is required in order to detect the properties of the complete data set. In this work, four distinct sampling schemes, namely the packet based uniform sampling, time-slot based uniform sampling, packet based random sampling and time-slot based random sampling are investigated from which packet flow length distributions are estimated and compared with the actual data. No major differences are observed amongst the strategies based on the analysed data.