<p>The current study aims at developing a regional frequency analysis for Maximum Annual Temperature (MAT) in the hydrographic basins of Colombia. The methodology that was implemented for the regionalization is the <em>L-Moments</em>, proposed by Hosking (1990). The work was performed in five stages: data analysis, computing of <em>L-Moments</em>, identification of homogeneous regions, selection of probability and distribution of the density function and estimation of quantities, and finally the construction of regionalized maps. Overall, fifteen (15) homogeneous regions were identified and selected for annual maximum temperature, which meet certain criteria of homogeneity and discordance. Several distributions for regional frequency analysis were tested in order to select the best probability function: the Z^DIST statistic was used, considering that the absolute value must be smaller than 1,64. Finally, temperature maps were generated for different return periods. Using the <em>L-Moments</em> methodology, the regionalization procedure was done using the annual average maximum temperature as the key scale parameter. However, the regionalization procedure was done using the geographic location; the physical parameters by gauging stations (Wallis, 2007), with stations located in the same catchment, and considering the coherent-basins approach.</p>