OBJECTIVES: The Zarit Burden Inventory (ZBI) is a 22-item self-report scale frequently used to asess patients’ caregiver burden on several dimensions. As a multidimensional instrument the interpretation of its total score is sometimes unclear. Our aim was to obtain a short-ZBI unidimensional scale based on Item Response Theory (IRT) approaches.  METHODS: The validation sample comprised 246 caregivers of patients diagnosed with dementia and recruited for an ongoing multi-center randomized clinical trial on the efficacy of psychoeducational interventions (EDUCA-2 trial). The pre-randomization 22-item ZBI was analyzed according to the Samejima’s graded response model to select the more informative items. The dimensionality of the scale was further tested with Confirmatory Factor Analysis (CFA). Finally, discriminant validity was assessed by Receiving Operator Characteristic (ROC) analysis and the Area Under the Curve (AUC) contrasting the short scale total score against the psychological distress criterion evaluated with the General Health Questionnaire 28-item at 5/6 cut-off. RESULTS: A 12-item short-ZBI was selected. It covered 87% of the total 22-item ZBI information and showed appropriate item curve characteristics according to the Samejima’s model. The short-ZBI had an internal reliability of 0.89 (Cronbach’s alpha), and was compatible with a unidimensional latent structure for the burden construct (CFI = 0.99; RMSEA = 0.05). According to the GHQ-28 cut-off 131 caregivers (53% of the total sample) could be considered at high risk for developping psychological distress. The discriminant validity of the short-ZBI scale against that criterion was good (AUC = 0.84, 95% CI = 0.79 to 0.89) and not significantly different from the parental 22-item ZBI (p = 0.85). CONCLUSIONS: We have found good psychometric properties for the short-ZBI scale derived from IRT. Its unidimensionality might be important to enhance its interpretation. Further psychometric studies, mainly on its sensitivity to change are now warranted.