Techniques are presented for estimating the magnitude and frequency of floods at ungaged sites on unregulated rural streams in New York, excluding Long Island. The discharge-frequency data and basin characteristics of 220 stream-gaging stations in New York and adjacent States were used in developing multiple linear regression equations for floods ranging in recurrence interval from 2 to 100 years. Separate equations were developed for northern, southeastern, and western New York. Standard errors of estimate of the 100-year flood range from 32.9 percent in the southeastern region to 42.8 percent in the western region. Drainage area is the independent variable needed in all equations; other variables needed, depending on region, are main-channel slope, storage index, and mean annual precipitation. A method is given for obtaining improved discharge-frequency relationships at gaged sites by weighting log-Pearson type III and regression estimates according to their variances. Basin characteristics, log-Pearson type III statistics, and regression and weighted estimates of the frequency-discharge relationship, are tabulated for the gaging stations used in the regression analyses. (Kosco-USGS)