The untimely calibration of electric energy metering equipment may result in some errors. It is of great significance to study measurement error detection methods. Therefore, a measurement error detection method for smart electricity meter based on maximum likelihood method and decision tree is designed. Data cleaning and dimensionality reduction are implemented on the raw measurement data of smart electricity meters. Based on the maximum likelihood method, the separability of abnormal feature data items after cleaning and dimensionality reduction is calculated. The higher the separability obtained from the calculation, the more obvious the classification effect of the data item on anomalies. The data items with higher separability are selected and input into the XGBoost algorithm to construct an accurate judgment function. The hyperparameter search is implemented through grid search method, combined with the adaptive ability and self-learning ability of algorithm, to confirm measurement errors, improve the accuracy judgment ability for electricity meter measurement errors, and achieve the measurement error detection of smart electricity meter. The test results show that this method has achieved high detection accuracy in multiple seed regions, and overall has good stability and generalization ability.