The primary purpose of this research is to compare and evaluate the effectiveness of observed score methods -Mantel-Haenszel, logistic regression- and latent score methods -IRT-LR, SIBTEST- which used to determine DIF under variety conditions. These methods were compared by simulation study. Sample sizes, ability distribution, proportion of items with DIF were considered for data simulation conditions. Results of this research revealed that latent score methods were more sensitive and effective in determining items with DIF rather than observed score methods. Latent score methods were more liberal and observed score methods were more conservative in identifying items with DIF. As a result, MH, SIBTEST and IRT-LR methods present consistent result in determining uniform DIF in all conditions. Furthermore, consistent results were found in identifying non-uniform DIF with LR, SIBTEST and IRT-LR methods.