Ship length is extremely useful for ship classification; therefore, if it is possible to derive accurate ship length estimates from ISAR (Inverse Synthetic Aperture Radar) data, the classification and identification problem becomes much simpler. This note demonstrates that it is possible to obtain extremely accurate measurements of ship length from ISAR images. The SAIC procedure used to produce ISAR images [Melendez and Bennett] includes ship length estimates for each frame. Robust length estimates based on 2000 frames are accurate within +/- 10.5, but we show that they can be improved significantly by the use of a frame selection procedure based on a neural network, which achieves an accuracy of +/- 2.3.