We examined the discussion processes through which two independent consensus-based review panels detected errors in the same sample of items from an international test translated from English to Chinese. The discussion processes were defined according to four events: (1) identifying a potential error; and (2) agreeing with, (3) disagreeing with, and (4) elaborating an opinion expressed by other panelists. We found that, while the two panels had similar error detection rates, only half of the errors detected by the two panels altogether were detected by both panels. In addition, of the errors detected by the two panels, more than half were detected by the panels through different discussion processes. No discussion process occurred substantially more frequently or less frequently for any translation error dimension. We conclude that the unique combination of backgrounds, skills, and communication styles of panel members and the unique combination of textual features in each item shape which errors each panel is capable of detecting. While panels can be highly effective in detecting errors, one single panel may not be sufficient to detect all possible errors in a given set of translated items. Consensus-based translation error review panels should not be assumed to be exchangeable.