BACKGROUND: The objective was to analyze locoregional (LR) failure patterns in patients with head-and-neck cancer (HNC) treated using intensity-modulated radiotherapy (IMRT) with whole salivary gland-sparing: parotid (PG), submandibular (SMG), and accessory salivary glands represented by the oral cavity (OC). METHODS: Seventy consecutive patients with Stage I-II (23%) or III/IV (77%) HNC treated by definitive IMRT were included. For all LR failure patients, the FDG-PET and CT scans documenting recurrence were rigidly registered to the initial treatment planning CT. Failure volumes (Vf) were delineated based on clinical, radiological, and histological data. The percentage of Vf covered by 95% of the prescription isodose (Vf-V95) was analyzed. Failures were classified as "in-field" if Vf--V95 >= 95%, "marginal" if 20% < Vf-V95 < 95%, and "out-of-field" if Vf-V95 <=20%. Correlation between Vf-V95 and mean doses (Dmean) in the PG, SMG, and OC was assessed using Spearman's rank-order correlation test. The salivary gland dose impact on the LR recurrence risk was assessed by Cox analysis. RESULTS: The median follow-up was 20 months (6--35). Contralateral and ipsilateral PGs were spared in 98% and 54% of patients, respectively, and contralateral and ipsilateral SMG in 26% and 7%, respectively. The OC was spared to a dose <=40 Gy in 26 patients (37%). The 2-year LR control rate was 76.5%. One recurrence was "marginal", and 12 were "in-field". No recurrence was observed in vicinity of spared structures. Vf-V95 was not significantly correlated with Dmean in PG, SMG, and OC. The LR recurrence risk was not increased by lower Dmean in the salivary glands, but by T (p = 0.04) and N stages (p = 0.03). CONCLUSION: Over 92% of LR failures occurred "in-field" within the high dose region when using IMRT with a whole salivary gland-sparing strategy. Sparing SMG and OC in addition to PG thus appears a safe strategy.