We determined subjective mutual intelligibility and linguistic similarity by presenting recordings of the same fable spoken in 15 Chinese dialects to naive listeners of the same set of dialects and asking them to rate the dialects along both subjective dimensions. We then regressed the ratings against objective structural measures (lexical similarity, phonological correspondence) for the same set of dialects. Our results show that subjective similarity is better predicted than subjective mutual intelligibility and that the relationship between objective and subjective measures is logarithmic. Best predicted was log-transformed subjective similarity with R2 = .64.