In wireless channel measurements, the relatively weak cross-polarized multipath components (MPCs) are typically severely affected by the measurement noise level. As shown in this letter, the typical cross-polarization ratio (XPR) model parameter estimation, which ignores the existence of censored samples, may lead to significant errors. We demonstrate how to achieve accurate parameter estimates with a maximum likelihood estimator that properly takes into account both the measured XPRs and the censored samples. Also, a new XPR model is presented in which the average XPR is modeled as a function of the MPC excess loss. The new model is shown to be insensitive to the channel measurement noise level. A practical example with measured data in an indoor environment at 60 GHz demonstrates the utility of the approach.