The classic Sudoku square design is applied for testing the null hypothesis when all observations in the data are precise and certain. In some cases, it is difficult for researchers to obtain exact observations due to circumstances beyond their control. Therefore, the observations may be imprecise and noted in intervals. We must therefore search for a suitable generalization of Sudoku square designs in uncertain environments. This paper aims to introduce the four various statistical models of the Sudoku square design under neutrosophic statistics, which provides a solution to the problem of imprecise and uncertain data. The test statistic of the proposed design per model is derived and applied for testing treatment effects. Based on the data analysis, and comparative study, it is concluded that the proposed neutrosophic Sudoku square design is informative, efficient, and adequate for use in the presence of uncertainty.