Regression analysis studies the relationships among the response variable and the predictors. Some common regression models include the linear regression model, the generalized linear model, the proportional hazards model, etc. After a data set has been collected, an important step in the regression analysis is to check the model assumption. A wrong model assumption can have a significant impact on the validity of the further inference. The existing tests to check model assumption must impose certain conditions on the joint cumulative distribution function of the response variable Y and the predictors Z or on the conditional expectation of Y on Z. If the imposed conditions are not satisfied, the asymptotic properties of the existing tests no longer hold and the test results are invalid. I propose a new model checking method, called the marginal distribution (MD) method. The MD method compares two marginal distribution functions: one is the marginal distribution function of the original response variable, the other is the marginal distribution function of a new variable which is constructed based on the model in the null hypothesis. The MD method consists of diagnostic plotting and hypothesis testing. One contribution of the MD method is that it does not need to impose any conditions on the joint cumulative distribution function or on the conditional expectation and thus it is alway valid. I present the simulation study on testing the common regression model assumption and the simulation results suggest that, when the existing methods are invalid, the MD test is still consistent. Moreover, when the existing tests are valid, the performance of the MD method is also satisfactory.