<h3>Purpose/Objective(s)</h3> Despite advancements in radiotherapy, approximately 30% of nasopharyngeal cancer (NPC) patients develop treatment failure. Dynamic on-treatment biomarkers may provide early evidence of response and improve risk stratification and treatment selection. Using daily cone beam computed tomography (CBCT) scans, we measure air volume recovery, an automated surrogate of tumor regression, and report its association with outcomes. <h3>Materials/Methods</h3> Patients with NPC treated with curable intent were included. Primary gross target volume (GTVp) was propagated from planning CTs to daily CBCTs by rigid registration and a density threshold was applied to measure the air volume in a 5 mm uniform expansion of the GTVp. Air volume was expressed as percent of planning GTVp volume. Missing values were interpolated, air volume trajectories were smoothed by sliding window averaging, and area under the curve (AUC) was calculated to provide a summary metric capturing both magnitude and rapidity of air volume recovery. Primary endpoint was recurrence free survival (RFS). <h3>Results</h3> Between 2013 and 2015, 39 consecutive NPC patients (2 stage II, 26 stage III, 11 stage IVA-B), ECOG 0-1, underwent radical radiotherapy (n=5) or chemoradiotherapy (CRT) (n=34) to a planned dose of 70 Gy in 35 fractions (Fx). Median AUC of the air volume recovery curve was 2.8 (Range: 0.05-7.4). Patients with above-median AUC had longer RFS (HR=0.28, log-rank p=0.045, median follow-up: 83 mo). Exploration of clinical factors associated with air recovery showed that AUC was higher in CRT than RT alone (OR: 1.74±1.60, p=0.04), lower in large tumors with high GTVp volume (OR: -0.037±0.18, p<0.001), and was not correlated with disease stage. Current smokers had 26% lower median AUC than never/former smokers (p=0.132). AUC was lower in patients with detectable end-of-treatment Epstein-Barr virus (EBV) circulating tumor DNA (median AUC 0.99 vs. 2.8 in undetectable EBV, p=0.021). We performed exploratory <i>post-hoc</i> analysis on air volume velocity (slope of air volume curve) and identified two independent putative positive prognostic biomarkers: high mid-treatment slope and negative end-treatment slope. Patients without failure had higher mid-treatment air recovery velocity (Fx 15-22), with Fx 18 velocity yielding the maximal RFS difference (HR: 0.12, p = 0.007). Negative velocity at the end of treatment was associated with improved regional (p=0.033) and distant metastatic (p=0.013) control. <h3>Conclusion</h3> Greater and more rapid tumor regression, inferred by the CBCT surrogate measure of air volume recovery, was associated with improved RFS. This discovery cohort also allowed us to identify promising dynamic biomarkers which might guide treatment adaptation. Specifically, we observed strong clinical outcome associations with mid- and end-treatment air recovery velocities. Validation of these biomarkers is ongoing in a large cohort of NPC patients with mature follow up.