The factorial structure of the Kessler Psychological Distress Scale 10-item (K10) and 6-item (K6) has been explored, with conflicting findings across studies. Bifactor modelling has been applied to the K10, but not to the K6, which is commonly used in clinical and research settings; hence, exploring its dimensionality can demonstrate the utility of its subscales. The aim was to determine, using confirmatory factor analyses (CFA), whether the bifactor model of K6 is a suitable representation of psychological distress among a community-dwelling sample of adults living with severe and persistent mental illness. Randomised controlled trial participants (n = 335) completed the K6 at baseline (median score = 9/24). CFA of the various models were conducted with diagonally weighted least squares, mean and variance adjusted. The unidimensional model was rejected. The fit statistics for the correlated model and the higher-order model were identical and demonstrated comparative fit index and Tucker-Lewis index within range; however, the RMSEA values were outside the target. In the correlated model, the correlation between the factors was very high. The bifactor model had the best fit for the data. Inspection of the reliability indices from the bifactor model showed that the way participants responded to the K6 items was far more influenced by a general psychological distress factor compared with anxiety and depression factors. Tests of measurement invariance showed that participants interpreted K6 items consistently between genders but not between younger vs older age groups. Further research across larger, diverse populations should explore the suitability of suggested instrument modifications.