Trends and uncertainty in radiosonde temperature records for six sample stations in eastern China are assessed. Results from a complex approach using metadata and a two-phase regression (M-TPR) to capture known and unknown metadata events respectively are compared with an ensemble of possible solutions generated by the Met Office automated homogenization system (QUARC). Independent satellite records from the Microwave Sounding Unit (MSU) record are used to validate breakpoints over the satellite era. Differences in the treatment of metadata and the strictness of the statistical breakpoint detection methods used lead to relatively poor agreement in breakpoint identification. Agreement in long-term (1958–2003) trends in the homogenized data was found to result from a fortuitous cancellation of large differences in the pre- and post-satellite era trends between the two approaches. A consideration of independent MSU satellite data lends some credence to the presence and calculated magnitude of many of the assigned breakpoints that were not associated with recorded metadata events, in the later part of the record. However, it also highlights that neither of the approaches is likely to be perfect at identifying breaks. Improved metadata are likely to prove vital in confirming the presence of these breaks and hence the veracity of the various homogenization approaches to data for eastern China