Constant false-alarm rate (CFAR) techniques are compared in an airborne pulse-Doppler detection application. Standard methods such as CA-CFAR, GO-CFAR and OS-CFAR prove to be inadequate in dealing with the extreme variations of power between radar cells, resulting in large CFAR losses. Two new techniques, nadir-blanked CFAR and gradient CFAR, were specifically designed to accommodate these variations. Nadir-blanked CFAR simply does not attempt detection in the first few gates that include or exceed the altitude range. Gradient CFAR does not perform as well as nadir blanking, but is still considerably better than standard CFAR methods. The method is similar to GO-CFAR, but the GO-threshold is boosted by the gradient between two groups of reference cells. Both methods allow for large reductions in the CFAR multiplying constant and therefore substantial improvements in detectability over the standard methods.