As an approach of granular computing, the sequential three-way decision(S3WD) model has been widely studied in practical applications. Fruitful results have been achieved in existing research on improving the accuracy of S3WD model. However, the positive and negative regions obtained by a pair of probability thresholds \( (\alpha ,\beta )\) introduced by the decision-theoretic rough set model will inevitably lead to the misclassification of some objects. To improve the accuracy of the S3WD model, this paper proposes a S3WD model from the perspective of uncertainty. Firstly, the uncertainty of the equivalence class is defined, and a pair of thresholds are constructed according to the shadowed set theory to measure the uncertainty of the equivalence class, and the equivalence class with large uncertainty in the positive and negative region near both sides of the boundary region is screened out. Finally, the S3WD model based on uncertainty measurement is proposed. The experimental results show that the proposed model has better performance in classification ability compared with the classical S3WD model.