It is well-known that phonemic analysis often has multiple solutions. For example, the Chinese word ‘fast’ have been transcribed as [khuai] and segmented as [k]-[h]-[u]-[a]-[i] in one analysis, and transcribed as [khwai] and segmented as [khw]-[a]-[i] in another, with quite a few other alternatives in between. Each analysis in turn will yield a different number of phones and phonemes. We propose a solution to the ambiguities, which makes use of two criteria. First, instead of simply comparing the number of phonemes in different analyses, we propose comparing the number of ‘basic phonemes’. Second, we compare how each analysis accounts for occurring and non-occurring syllables against all possible combinations of phonemes. We demonstrate our approach with two segmentation options of Old Chengdu, the ‘finest’ segmentation vs. the ‘coarsest’ segmentation. We show that they end up being essentially the same analysis.