Window analysis basically assumes that no technical change has taken place in the periods under review. This underlying assumption has been neglected in most research works. In the current study, using non-parametric statistical Wilcoxon test and Friedman test between two time periods and over two time periods, the shift frontier between the periods under review has been investigated and determined. Accordingly, the length of the windows was selected in the way that the basic assumption of window analysis was observed. In this research, the data of 7 distribution-channels of Pars Khazar Company was picked up in time periods of one year (2005-2011). After running the statistical tests, it was revealed that shift frontiers had occurred within the periods (2007 and 2008), (2008 and 2009) and (2010 and 2010). So four analyses with the lengths of 3, 1, 2, and 1 were performed in which the average performance of the units was obtained as 0.832, 0.987, 0.906 and 0.953, respectively.