The health index (HI) of machine condition must be sensitive and robust in complex working conditions. A systematic HI will assess machine performance automatically, reliably, and in a timely manner without intervention. This paper proposes a subspace clustering HI in a model using reference data on component health. Unlike the conventional HIs empirically learned from raw feature sets, a subspace clustering HI aims to automatically describe the migration and variation of the condition clustering distribution in a series of two-class subspace models derived from the raw data. First, in a featureless process, a covariance-driven Hankel matrix is directly constructed from the raw time-domain signal, and principal component analysis is used to separate the feature subspace and noise null-space. Second, in the index construction process, the reference health subspace data (from healthy data) and the monitored subspace data (from monitored data) are combined to construct a referenced model. Thus, a new spatial clustering HI with kernel operation is implemented to assess the current bearing performance and reveal discriminative features. The effectiveness of the proposed subspace clustering HI for the detection of abnormal condition is evaluated experimentally on bearing test-beds, using a mobile mapping mode. A novel subspace clustering chart, CUSUM-based spatial clustering HI, is developed to depict the real bearing performance degradation. Compared to the regular HI (e.g., root mean square), the proposed approach provides a more accurate and reliable degradation assessment profile with an early fault occurrence alarm. The experimental results show the potential of the proposed spatial clustering analysis to assess bearing degradation.