In the process of rolling of Quenched and Tempered (Q&T) reinforcement bars, it is required for bars to obtain both good strength and elongation properties. Many parameters associated with the raw material and process affect the final properties of bars. In industry, destructive tests cannot be performed frequently to measure the final properties of bars. This study is an attempt to investigate the usefulness of various computational methods in the design of data based quality prediction models for Q&T reinforcement bars. To work with the problem, twelve parameters related to the raw material and process were selected as inputs and three features as outputs. Then, multiple linear regression, principal components regression, partial least squares regression, artificial neural networks and locally weighted regression were sequentially implemented to construct prediction models. In advance, to increase the prediction quality, we have proposed a selective approach which has three different modules. Two of the three select the proper models through pre-screening the process in terms of the similarity of the query to the pre-determined clusters. The third approach tries to incorporate the local correction from the value predicted by global linear model. For the collected data from two real industries, the proposed selective approach outperformed the five individual prediction models in overall comparison.