Hoarding is a mental-health problem manifested by excessively saving items irrespective of their value. One factor considered in the assessment of hoarding severity is the amount of clutter in a dwelling, usually quantified through a visual scale called the “Clutter Image Rating” (CIR). This requires a visit to an individual's home and rating clutter on the CIR scale, a time-consuming, subjective, and often non-repeatable process. To date, several methods were proposed for automatic rating of clutter from images but were evaluated on relatively narrow and unbalanced datasets. In this paper, we introduce a new 1,800-image, balanced dataset of clutter images that has been CIRrated by health professionals. We also propose a new method for rating clutter that is based on the Vision Transformer. We evaluate the proposed method against a state-of-the-art clutterrating method on two datasets via 4-fold cross-validation. We also perform two ablation studies (loss-function parametrization and data augmentation). In quantitative comparisons, we measure accuracy and accuracy within ±1 since even health and humanservice professionals admit to challenges in assigning exact CIR values. The proposed method is shown to outperform the best method to-date by 4.50-7.12% points in exact CIR matching and by 5.80-6.53% points in matching with a slack of ±1. Even more importantly, the new method achieves accuracy of over 93% with a slack of ±1 suggesting it can be a reasonable proxy for ratings by health professionals and a valuable tool in the assessment and treatment of hoarding disorder.