This study investigates the psychometric properties of Chinese-version short-form 10-item CES-D (i.e., CESD-10) in clinical depression patients and non-clinical college students by structural equation modeling approach. Several statistical procedures are applied to investigate the psychometric properties of CESD-10. First, the single factor model of CESD-10 shows merely mediocre fit due to “error term correlations”. The two pairs of error term correlations indicate that CESD-10 has redundant items and negative-worded effect. The revised model, based on the modification index, demonstrates good fit, and is adopted for subsequent analysis. Results of the SEM-based method show that configure invariance holds, while weak factorial invariance is not supported. A comparison the psychometric properties of the two groups shows that the clinical sample has better reliability, while the non-clinical sample has superior factorial validity. This finding confirms that the Chinese-version CESD-10 is most appropriate for the nonclinical, general population, although it is also valid for the clinically depressed.