In generalized linear model, we always need to check the model assumption and there exist several methods, but existing plots and tests require certain constraints on the joint cumulative distribution function of the response and covariate, I will describe the methods shortly, the paper proposed a new method to test the model assumption, by comparing the non-parametric MLE and the estimator based on the null hypothesis, which is called the marginal distribution method, several simulation results will be provided and we can see this new method can give us satisfactory results.