Language impairment is a sensitive biomarker for the detection of cognitive decline associated with mild cognitive impairment (MCI). Recently, knowledge about distinctive linguistic features identifying language deficits in MCI has progressively been enriched and accumulated. However, the employment of a single speech task to elicit connected speech (e.g., structured vs. spontaneous conversations) might limit the generalization of salient linguistic features associated with MCI. Not to mention the scarcity of reports on analysis of extended speech of Chinese. The present study aimed to examine if connected speech production in both situational picture description and spontaneous self-introduction tasks could be used to distinguish individuals with psychometric evidence of MCI and those who were cognitively intact. Speech samples produced by 75 elderly native speakers of Mandarin Chinese, including 19 with MCI and 56 healthy controls were obtained. Macrostructural aspects of language, including lexico-semantic, syntactic, speech fluency, and acoustics were analyzed by applying the linear mixed-effect regression model. Our study revealed decreasing linear trends in semantic contents and syntactic complexity, as well as significantly greater signs of disfluency and reduced speech production in participants with MCI. The findings extended what was reported in the literature, and carry important implications to the screening and diagnosis of suspected MCI.