Deep learning is the main focus of the machine learning field. However, it is noticeable that some researchers utilize broad learning to efficiently solve some problems in this field. Under such conditions, even though deep learning has obtained some breakthroughs, a comprehensive learning system is still valuable. Comparing the advantages and disadvantages of these two methods is difficult to achieve a general conclusion. However, outcomes about the performances of models are possible for a particular problem or data set. In this article, all the experiments are based on MNIST data set, using Multilayer Perceptron models to test the differences of broadness and depth. Nevertheless, the result of experiments shows it is tough to choose the best from these models. This may derive from the complexity of model structures.