Cross-polarization isolation is one of the key engineering parameters for a polarimetric radar system. Previous studies focused more on the calibration of cross-talk contamination. This paper presents a numerical evaluation of the requirement for cross-polarization isolation from the data users' perspective, i.e., the quantitative impact of polarization cross talk on polarimetric target decomposition and the associated applications such as classification and detection. Sensitivity analyses of several commonly used target decomposition parameters suggest that a theoretical lower bound of À32 dB isolation level is preferred to avoid any significant impact on these parameters. Our analyses with both simulated and real synthetic aperture radar (SAR) data show that a level of À25 dB would be acceptable for general terrain surface classification. This requirement is also true for man-made target detection application. Using simulated SAR images of man-made targets in natural environment, sensitivity analyses on two polarimetric detectors, Yang and Marino, both suggest that target detection performance would break down rapidly if isolation deteriorates from À25 dB to À20 dB.