1: \begin{abstract}

2: This paper studies model checking for general parametric regression models that have no dimension reduction structures on the high-dimensional vector of predictors. Using an existing U-statistic type test as an initial test, this paper combines the sample-splitting technique and the conditional studentization approach to construct a COnditionally Studentized Test (COST). Its significant feature is that whether the initial test is global or local smoothing-based, and whether the dimension of the predictor vector and the number of parameters are fixed or diverge at a certain rate as the sample size goes to infinity, the proposed test always has a normal weak limit under

3: the null hypothesis. Further, the test can detect the local alternatives distinct from the null hypothesis at the fastest possible rate of convergence in hypothesis testing.

4: We also discuss the optimal sample splitting in power performance. The numerical studies offer information on its merits and limitations in finite sample cases. As a generic methodology, it could be applied to other testing problems. 

5: 	\end{abstract}