Model selection is a major concern for many sciences and there have been a plethora of studies on model selection methods. Experts from different scientific fields have adopted different model selection procedures to select an appropriate model. Identifying the best subset among a large number of variables is the hardest part of model selection. There exist many methods for variable selection, and different methods select a different subset of variables. We can test their relative performance by comparing them. In the current study, we have compared the latest variables selection criteria from different classes of variables selection procedures; Autometrics, Elastic Net, and Extreme Bound Analysis (EBA). We have analyzed different situations (different data generating processes) by changing the combination of relevant variables, irrelevant variables, orthogonal variables, relevant variables are mutually correlated, irrelevant variables are correlated to relevant variables, relevant variables are serially correlated, heteroscedastic error terms and errors are auto correlated. When the variables are mutually correlated then Extreme Bound Analysis performs better than other model selection procedures. When the variance of the error term is heteroscedastic, the Autometrics performs superior to other rival model selection criteria. Similarly, Autometrics performs better in the case of orthogonal variables. For real data analysis, we use the data related to economic growth and its determinants for 32 countries. For in sample comparison, we use root mean square error (RMSE) and mean square prediction error (MSPE). Extreme Bound Analysis presents a superior predictive performance in terms of the lowest RMSE and MPSE that are 1.03 and 0.05 respectively. Supervisor:- Dr. Saud Ahmed Khan Co-supervisor:- Dr. Atiq Ur Rehman