Dimensionless precipitation-frequency curves for estimating precipitation depths having large recurrence intervals were developed for 2-, 6-, and 24-hour storm durations for three homogeneous regions in Montana. Within each homogeneous region, at-site annual precipitation maxima were made dimensionless by dividing by the at-site mean and grouped so that a single frequency curve would be applicable for each duration. L-moment statistics were used to help define the homogeneous regions and to develop the dimensionless precipitation- frequency curves. Data from 459 precipitation stations were used after application of statistical tests to ensure that the data were not serially correlated and were stationary over the general period of data collection (1900-92). The data were found to have a small, but significant, degree of interstation correlation. The GEV distribution was used to construct dimensionless frequency curves of annual precipitation maxima for each duration within each region. Each dimensionless frequency curve was considered to be reliable for recurrence intervals up to the effective record length. Because of significant, though small, interstation correlation in all regions for all durations, and because the selected regions exhibited some heterogeneity, the effective record length was considered to be less than the total number of station-years of data. The effective record length for each duration in each region was estimated using a graphical method and found to range from 500 years for 6-hour duration data in Region 2 to 5,100 years for 24-hour duration data in Region 3.