This study tries to present various patterns in lexical substitution errors from naturally-occurring speech-error corpus and from left-brain damaged patients in Mandarin. Using the error data to test some linguistic hypotheses is a long tradition, and these errors have been taken as important evidence regarding the units and structures involved in the cognitive representation of language. Recently, a few reports in the literature have started to examine some issues of the representation in Mandarin by looking at these speech-error patterns (e.g., Chen & Dell, 2003, 2006a, 2006b; Chen, Lin & Ferrand, 2003; Wan, 2006, 2007a, 2007b, 2007c; Wan & Jaeger, 1998, 2003). It is found that in English and Germanic languages, the incorrectly selected words in speech errors are almost always the same part of speech as their target words, and the lexical substitution errors often involve semantically related associates (e.g., Fromkin,1973; Dell & Gordon, 2003; Dell et al., 2004; Dell et al., 2007; Baron et al., 2008; Oppenheim & Dell, 2008). This is not a surprising fact that the lexical errors share grammatical categories with their targets since the grammatical patterns of the specific phrase impose important restrictions on the selection of words. In addition, those researchers also observed phonologically related lexical selection errors in their speech-error corpus. They found that the target-error pair in lexical substitutions shows some phonological associations in terms of initial segments, number of syllables, vowels/rhymes, and suprasegmental patterns. Therefore, the purpose of this study is twofold. First, the researcher will catalog all the various influences which may be factors in causing errors involving whole lexical items and rank-order them in terms of their importance. Second, these various influences and the rank-ordering will both be explained in terms of the content and organization of the lexicon. A total of 421 relevant lexical errors from native speakers of Mandarin and 286 errors from Mandarin aphasic patients, both collected by the author and the research team in a naturalistic setting, are provided to present the following findings. 1) Semantically-related errors could be classified in terms of the semantic relationship between the target and intended utterance. According to the five criteria (antonyms, hyponyms, near-synonyms, coordinates and associates), the rank-ordering for the normal subjects is the following: Hyponyms (32%) > Near-synonyms (25%) > Associations (21%) > Antonyms (12%) > Coordinates (10%); however, for aphasic patients, the data do not show the same rank-ordering. 2) Lexical errors that also share phonological similarities are found in normal subjects and aphasic patients. The target-error pair in lexical substitutions shows some phonological associations in terms of initial segments (25% for normals, 34% for aphasics), syllables (29% for normals, 34% for aphasics), tones (28% for normals, 25% for aphasics), but in both normal and aphasic patients, the target-error pair sharing the same vowel (1% for normals, 1% for aphasics) or rhyme (17% for normals, 6% for aphasics) is not quite common. This suggests that the incorrectly selected words and their intended words could share semantic relationship as well as phonological relationship. Some cases also show that the target-error patterns bear phonological relationship alone without any semantic association, and cases like this show that the initialness, syllables and tone effects are all prominent. Evidence from normal speakers’ slips and aphasic patients’ errors shows that the two interacting words tend to have a semantic relationship, a phonological relationship, or the semantic plus phonological relationship, suggesting some phonological facilitation in the Mandarin lexicon.