As an estimate of a location parameter for a given data set, $alpha$-trimmed mean has been studied for a long time by many statisticians because of its nice propoerties including robustness. However, its performance depends on the proportion of trimming say $alpha$. In this paper, we suggest a data-driven choice of $alpha$ and study its validity. Also, we suggest a new estimator and consider double-bootstrap to improve its performance. By using simulation study, the proposed method is compared with the exiting one in various cases. Real data sets are also analyzed by using the proposed method.