Conventional distributed-target-based polarimetric calibration algorithms estimate polarimetric distortions by assuming that the measured spatially averaged covariance matrix takes a specific form. However, when the underlying surface contains targets that do not satisfy the assumptions employed by those algorithms, the averaged covariance matrix may deviate from the desired form. As a result, poor estimates of distortion parameters may yield. It is known that spherically truncated covariance matrix is robust to outliers. Thus, we introduce it to the polarimetric SAR calibration routine. Experiment results on the airborne SAR data confirm that this method can effectively reduce the uncertainty of distortion estimates, hence improve the robustness of the calibration.