Background To develop a risk model based on dosimetric metrics to predict local recurrence in nasopharyngeal carcinoma (NPC) patients treated with intensive modulated radiation therapy (IMRT). Methods 493 consecutive patients were included, among whom 44 were with local recurrence. One-to-two propensity score matching (PSM) was used to balance variables between recurrent and non-recurrent groups. Dosimetric metrics were extracted, and critical dosimetric predictors of local recurrence were identified by Cox regression model. Moreover, recurrent sites and patterns were examined by transferring the recurrent tumor to the pretreatment planning computed tomography. Results After PSM, 44 recurrent and 88 non-recurrent patients were used for dosimetric analysis. The univariate analysis showed that eight dosimetric metrics and homogeneity index were significantly associated with local recurrence. The risk model integrating D5 and D95 achieved a C-index of 0.706 for predicting 3-year local recurrence free survival (LRFS). By grouping patients using median value of risk score, patients with risk score ˃ 0.885 had significantly lower 3-year LRFS (66.2% vs. 86.4%, p = 0.023). As for recurrent features, the proportion of relapse in nasopharynx cavity, clivus, and pterygopalatine fossa was 61.4%, 52.3%, and 40.9%, respectively; and in field, marginal, and outside field recurrence constituted 68.2%, 20.5% and 11.3% of total recurrence, respectively. Conclusions The current study developed a novel risk model that could effectively predict the LRFS in NPC patients. Additionally, nasopharynx cavity, clivus, and pterygopalatine fossa were common recurrent sites and in field recurrence remained the major failure pattern of NPC in the IMRT era.