The physiographical and climatological attributes of basins and maximum annual flood statistics were obtained from 38 hydrometric stations of Mazandaran province with minimum and maximum statistical periods of 13 and 56 years. The first, homogenous regions based on neighborhood approach were determined using canonical kriging method. The Canonical Correlation Analysis was used to design physiographic space using the six physiographic variables affecting the flood and two hydrological variables. The Gaussian model had the best fit to semi-variogram model in 10, 20, 50 and 100 year return periods, and regional estimates obtained using ordinary Kriging technique based on physiography space. The accuracy of results was evaluated using L-moments method for regional flood frequency analysis. In this method, Ward hierarchical clustering has been used to the determination of homogeneous regions. Homogeneity areas were adjusted by using homogenous and discordancy tests. Using goodness-of-fit Z ^DIST the Generalized Logistic distribution was selected for all clusters as the best regional distribution and regional estimates obtained by parameters selected distribution. Finally, performance methods evaluated using Jack Knife procedure and the five statistical indexes BIAS, BIASr, RMSE, RMSEr and NASH. Both methods give the same and desirable performance based on NASH, however results of other indexes showed that canonical kriging method provide better performance than linear moments.  Increasing the return periods will improve the estimation, but the L-moments offers better performance for short return periods.