Records of absenteeism from primary schools are valuable data for infectious diseases surveillance. However, the analysis of the absenteeism is complicated by the data features of clustering at zero, non-independence and overdispersion. This study aimed to generate an appropriate model to handle the absenteeism data collected in a European Commission granted project for infectious disease surveillance in rural China and to evaluate the validity and timeliness of the resulting model for early warnings of infectious disease outbreak. Four steps were taken: (1) building a ‘well-fitting’ model by the zero-inflated Poisson model with random effects (ZIP-RE) using the absenteeism data from the first implementation year; (2) applying the resulting model to predict the ‘expected’ number of absenteeism events in the second implementation year; (3) computing the differences between the observations and the expected values (O–E values) to generate an alternative series of data; (4) evaluating the early warning validity and timeliness of the observational data and model-based O–E values via the EARS-3C algorithms with regard to the detection of real cluster events. The results indicate that ZIP-RE and its corresponding O–E values could improve the detection of aberrations, reduce the false-positive signals and are applicable to the zero-inflated data.