Knowing the protein structures is essential in understanding the job descriptions of proteins involved in vital functions, drug design, and many more. On the other hand, protein structure prediction is an alternative bioinformatics sub-study field to shorten the process that takes a long time in the laboratory environment. Performance analyzes of the methods developed in this field are generally made on benchmark datasets. The size of the datasets directly affects the algorithm runtime. In this study, how to benchmark datasets are reflected in the results is analyzed. Within the scope of the study, two different benchmark datasets, CB513 and EVASet, and two different protein structure prediction methods, JPred and Porter, were used. The study is a source of inspiration for further studies with the idea of developing benchmark datasets that are comprehensive in terms of protein properties but contain as little data as possible in terms of data size.