Logistic regression models are widely used in medical diagnosis and prognosis. After the model is established from a training set, it is important to externally validate it using a validation data set for potential changes between the population that the training set is from and that the validation set is from. Currently, Cox’s test combined with some recalibration and model revision techniques has been widely used. In this talk, we discuss an alternative approach based on a novel idea of testing homogeneity. Based on our proposed tests, in the model recalibration stage we can determine when to combine the training and validation data, which can not be decided from Cox’s test. With this new feature, the performance of our updated model outperforms that from existing methods based on Cox’s test. This advantage is demonstrated in the simulation study and the real data study.