One way is to compare like instruments.  This figure shows correlation plots of AMSU TLS results obtained by two different groups (UAH and RSS) using exactly the same data from Dec 2004.  Results differ between the two groups because they employ different methods of correcting for presumed measurement errors.  The overall RMS discrepancy is ~1.4 K and differences exceed 4 K.  This illustrates the fundamental challenge of calibrating radiometers and indicates the basic level of reliability of the AMSU temperature data.  (S. Ho, NCAR, 2007)