Most current methods on polarization targets detection assume the target echoes to be ascertained in signal model during the detection period. But in fact, the target echo may obey a kind of distribution and the distribution relates with the polarizations. Aiming at improv- ing the performance of polarization detection, an adaptive generalized likelihood ratio (AGLR) polarization detector is proposed. The detector assumes the target echo to be a non-zero mean Gaussian distribution with respect to the polarization states of transceiver. It adaptively adjusts the polarization states of transceiver according to the variation of the polarization characteris- tics of target and clutter to maximize signal to clutter and noise ratio (SCNR). The problem of maximizing the SCNR is converted to a problem of maximizing a generalized Rayleigh quotient (GRQ). Based on the GRQ, the AGLR detector is proposed. The performance of the AGLR de- tector is theoretically analyzed and numerically validated through some numerical experiments. Compared with two conventional GLR detectors, the AGLR detector achieves a 3 dB to 10 dB improvement.