It is a great challenge to predict a protein structure and this challenge has fascinated researchers in different disciplines for many years. Basically the prediction process mainly includes two steps. With the first step that the generation of prediction model increasing fast, the second step that the quality estimation of predicted model i.e. identification of models’ native like structure becomes more and more important. In this study, we developed a simple and effective approach to identify the native-like protein structures among a set of decoys. Three different average measures were used in our study as follows: the average rmsd (armsd) , the average alignment score (AAS) and MAXSUB. This approach was evaluated by decoy set (Park-Levitt). Comparison of model quality revealed that a significant correlation existed between these parameters. For example, the average measure could be effectively used to identify native-like protein models. The performance of both armsd and AAS was better than that of clustering. Since many other measures could be used to assess the similarity between protein structures, other analogous approaches might be also useful for the identification of native-like proteins. Finally, data showed that its performance was better than that of other servers in predicting the targets in CASP6, CASP7, CASP9 and CASP10.