This paper compares variant characters in large-scale dictionaries from the pre-modern period with actual writing habits using a special subset of variants known as ‘semantic compounds’ (huiyi會意) as a case study. The results show that despite their prominent presence in traditional dictionaries, only a fraction of such variants were in everyday use. Most of the forms recorded in dictionaries were preserved and handed down as part of the lexicographic tradition, to some extent irrespective of genuine writing habits. Going one step beyond recognising that only some of the documented forms were at any given time in common use, the analysis presented here measures the discrepancy between dictionaries and manuscripts as a percentage.