In this paper, we use a combination of physiological and behavioral metrics of anxiety to detect changes in anxiety status in clinically anxious participants compared with healthy controls, which is important for intervening in a timely manner for the effective management of anxiety. Specifically, we first operationalize four phases of anxiety and select multimodal-multisensor feature candidates to assess those phases, considering preliminary results obtained in prior research employing a generalized mixed additive model-based analysis. Then, we evaluate the performance of selected features and their combinations in detecting the presence of anxiety phases, using the “Anxiety Phases” dataset. The results demonstrate that unimodal features of skin conductance response rate and mean rigidity of wrists detected all four temporal phases in ~50% of high-anxiety individuals. These two features detected at least three phases in ~90% of high-anxiety individuals. The fusion of these two features with an additional postural feature detected all four temporal phases in 65% of high-anxiety individuals. The multimodal-multisensor combination of these three features represented a 30% improvement compared with the best unimodal predictive feature. Implications of these findings are discussed for developing accurate real-time multimodal-multisensor anxiety management systems for clinical populations.